# q-bio.BM Published articles for q-bio.BM. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Static Recovery Is Not Dynamic Stability: Dynamics-Aware Benchmarking of Protein Motif Scaffolding DevFeed: [Static Recovery Is Not Dynamic Stability: Dynamics-Aware Benchmarking of Protein Motif Scaffolding]() Original publisher: [Read original article]() Author: Emile de Bruyn, Moritz Sch\\"affler, Jannik Schneider, Sebastian R. Schmidt, Chanwoong Hwang, Aleena Siji, Utkarsh Upadhyay, Stefan Bauer, Alexander Schug, Stefan Kesselheim, Artur Yakimovich, Alina Ba Published: 2026-11-06T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Molecular Dynamics](), [Computer science]() Tags: [benchmarking](), [cs-lg](), [evaluation](), [machine-learning](), [molecular-dynamics](), [q-bio-bm](), [recovery](), [static]() ### AI overview The benchmark finds that static motif recovery does not establish dynamical stability of the full protein scaffold. BILBY-PRO expands evaluation to 220 cases from 174 PROSITE motif-pattern entries. Across six scaffold-generation models, RFDiffusion and Proteina solve more targets, while ESM3 yields more unique successful scaffolds on the targets it solves. BILBY-MD compares statically solved MotifBench designs with reference molecular-dynamics simulations using motif and scaffold drift, motif-contact preservation, and motif-dihedral free-energy landscapes. Of 55 statically solved model-problem pairs, 31 meet the MD-supported criterion and 8 pass all checks. Compact, self-contained motifs are often supported; interface-like or context-dependent motifs fail more frequently. The authors frame evaluation around target coverage, unique-solution yield, and MD-supported dynamical fidelity. ### Source excerpt arXiv:2611.07723v1 Announce Type: new Abstract: Protein motif scaffolding benchmarks assess whether generative models can build scaffolds around functionally important residues. Their conclusions depend on which motifs are tested and which criteria define success. Existing benchmarks typically use small hand-curated motif sets and static refolding criteria, but protein function often depends on conformational ensembles and transitions. Here, we address both benchmark scope and success criteria. We introduce BILBY-PRO, a systematically derived benchmark of structurally conserved functional motifs from PROSITE-linked experimental structures, comprising 220 cases from 174 PROSITE motif-pattern entries. Evaluating six recent scaffold-generation models reveals distinct patterns in target coverage and unique-solution yield: RFDiffusion and Proteina solve more targets, whereas ESM3 produces more unique successful scaffolds on the targets it solves. We further introduce BILBY-MD, a reference-based all-atom molecular dynamics evaluation framework. Applied to statically solved designs from the original MotifBench benchmark, it compares motif drift, scaffold drift, motif-contact preservation, and motif-dihedral free-energy landscapes against reference simulations. Of 55 statically solved model-problem pairs, 31 satisfy our MD-supported criterion and only 8 pass all checks. Compact, self-contained motifs are often MD-supported, whereas interface-like or context-dependent motifs fail more frequently. Static refolding success therefore provides useful evidence of local motif recovery but does not establish dynamical fidelity of the full scaffold. BILBY highlights three complementary axes of motif-scaffolding evaluation: target coverage, unique-solution yield, and MD-supported dynamical fidelity. ## Geometric Relevance Is Not Sufficient: Evaluating Focused Post-Training Supervision in Diffusion Docking DevFeed: [Geometric Relevance Is Not Sufficient: Evaluating Focused Post-Training Supervision in Diffusion Docking]() Original publisher: [Read original article]() Author: Michael Brown, Maryam Rahimimovassagh, Mehdi Yazdani-Jahromi, Sanjeda Sara Jennifer, Ivan Garibay Published: 2026-11-06T05:00:00Z Content type: news Language: en Sources: [arXiv: Machine Learning]() Topics: [Model Development](), [post-training]() Tags: [cs-lg](), [diffusion](), [evaluation](), [performance](), [post-training](), [q-bio-bm](), [recovery](), [sampling](), [supervision](), [training]() ### AI overview In a fixed-budget DiffDock study, focusing post-training on torsional corrections with greater Cartesian consequences or more consistent conditional targets did not consistently improve docking success over broad continuation. At the <5 Å threshold, broad continuation achieved the highest mean Top-1 success among trained policies. The starting model retained better fixed score fidelity, while broad continuation had higher mean Top-1 success with the same aggregate best-of-10 recovery. The results distinguish score fidelity, pose quality, sampling capacity, and confidence-selected performance; they do not establish an advantage for focused or mixed supervision under the tested budget. ### Source excerpt arXiv:2611.08249v1 Announce Type: new Abstract: Diffusion-based molecular docking requires learning a corrective score field across a broad corruption distribution, yet individual pose errors can differ substantially in their structural consequences. This creates a post-training conflict between preserving broad score-field coverage and concentrating supervision on corrections that appear more relevant to docking geometry. Using DiffDock, a fixed-budget study compares broad continuation with torsional supervision focused on Cartesian consequence or conditional target consistency, and with a broad/focused mixture. Evaluation separates fixed score fidelity, generated-pose quality, best-of-10 sampling capacity, and confidence-selected Top-1 docking performance. At <5 Angstrom, broad continuation had the highest mean Top-1 success among trained policies, and focused or mixed supervision showed no consistent advantage under the tested budget. The starting model had better fixed score fidelity, whereas the evaluated broad-continuation model had higher mean Top-1 success with the same aggregate best-of-10 recovery. Focusing supervision on structurally consequential or more consistently specified corrections did not consistently improve returned-pose success over broad continuation. ## A Comprehensive Benchmark of Computational Models for Antibody--Antigen Binding Affinity DevFeed: [A Comprehensive Benchmark of Computational Models for Antibody--Antigen Binding Affinity]() Original publisher: [Read original article]() Author: Xinyan Zhao, Yi-Ching Tang, Victor J Cantu, Junseok Lee, Adam E Stogsdill, Ibraheem M Hamdi, Ashwin Kumar Ramesh, Zhiqiang An, Xiaoqian Jiang, Yejin Kim Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence](), [arXiv: Machine Learning]() Topics: [Model Development](), [Benchmark](), [Fine-tuning]() Tags: [abstract](), [affinity](), [all](), [arxiv](), [benchmark](), [benchmarks](), [case-study](), [cdr](), [compare](), [complex](), [cs-ai](), [cs-ce](), [cs-lg](), [design](), [developing](), [evaluation](), [experimental](), [fine](), [fine-tuning](), [folding](), [framework](), [function](), [models](), [q-bio-bm](), [q-bio-qm]() ### AI overview AbBiBench evaluates antibody-antigen complexes against 216,560 experimental measurements across 13 antibodies and 9 antigens, comparing 20 computational approaches without task-specific training or fine-tuning. Global structure-conditioned inverse folding produced the strongest and most consistent affinity rankings; antibody-specific fine-tuning did not reliably improve them. The rankings were more useful for prioritizing likely affinity gains than for ruling out weak variants. In an H1N1 case study, evaluating F045-092 CDR-H3 variants against both predicted binding potential and evolutionary sequence plausibility exposed a trade-off; all 21 selected candidates showed H1-reactive ELISA signals. ### Source excerpt arXiv:2506.04235v3 Announce Type: replace-cross Abstract: Computational antibody design requires benchmarks that measure functional binding rather than similarity to natural sequences. Here we present AbBiBench, which treats the antibody--antigen complex as the evaluation unit and harmonizes 216,560 experimental measurements spanning 13 antibodies and 9 antigens. We compare 20 computational approaches without task-specific training or fine-tuning. The analyses show that global structure-conditioned inverse folding yields the strongest and most consistent affinity rankings, whereas antibody-specific fine-tuning does not consistently improve performance. Across models, rankings are more useful for prioritizing likely affinity gains than for excluding weak variants. In a subsequent case study, multi-objective evaluation of model-generated F045-092 CDR-H3 variants for H1N1 binding revealed a trade-off between predicted binding potential and evolutionary sequence plausibility; all 21 selected candidates produced H1-reactive ELISA signals. AbBiBench provides an open framework for developing and evaluating function-aware antibody design methods. ## iGC: Antibody Optimization via Reinforcement Learning over a Learned Somatic Hypermutation Prior DevFeed: [iGC: Antibody Optimization via Reinforcement Learning over a Learned Somatic Hypermutation Prior]() Original publisher: [Read original article]() Author: Thao Nguyen, Robert G. Alberstein, Ewa Nowara, Homa MohammadiPeyhani, Holly R. Steach, Sarah A. Robinson, Yuval Shechter, Donald W. Lee, Jason A. Vander Heiden, Franziska Seeger, Andrew Leaver-Fay, Ri Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Reinforcement learning](), [Machine learning](), [Model Development](), [Operations research and optimization]() Tags: [affinity](), [cs-lg](), [diversity](), [optimization](), [prediction](), [q-bio-bm](), [reinforcement-learning](), [sequence]() ### AI overview iGC combines biologically plausible somatic hypermutation with antigen-specific antibody optimization in a two-stage reinforcement-learning framework. It first learns a mutation prior from B-cell lineage parent-child pairs and adapts it to a target affinity-maturation dataset; then it initializes a policy from that prior and uses KL regularization to keep optimization near the frozen prior. A Reward-Tax Meta-Gradient objective pairs RewardNet, which smooths the affinity landscape, with TaxNet, which penalizes reward overexploitation such as repetitive substitutions and tolerated framework mutations. The abstract reports that the mutation prior more than doubles the strongest baseline in mutation-site prediction, and that optimization maintains SHM sequence structure while achieving competitive reward-diversity products across candidate-selection cutoffs. ### Source excerpt arXiv:2611.05270v1 Announce Type: new Abstract: Antibody affinity maturation is a natural optimization process in which B cells acquire mutations through somatic hypermutation (SHM), undergo selection for improved antigen binding, and iteratively evolve toward higher affinity. Computationally modeling this process requires capturing both biologically plausible mutations and antigen-specific selection. However, existing mutation models do not directly optimize binding, while affinity optimization often neglects realistic mutation trajectories. We introduce iGC (in silico Germinal Centers), a two-stage reinforcement learning (RL) framework for antigen-specific antibody optimization. Stage 1 learns an SHM mutation prior from parent-child pairs in B-cell lineage trees and adapts it to a target affinity-maturation dataset. Stage 2 initializes an RL policy from this prior and optimizes antigen binding with Kullback-Leibler (KL) regularization toward the frozen prior. To prevent reward exploitation, including repetitive high-reward substitutions and structurally tolerated framework mutations, we introduce a Reward-Tax Meta-Gradient objective. RewardNet smooths the affinity landscape, while TaxNet penalizes reward overexploitation, jointly defining a learned meta-reward for policy optimization. We evaluate iGC on mutation prediction, SHM consistency, affinity optimization, and mutation diversity. The mutation prior more than doubles the strongest baseline in mutation-site prediction. Following RL optimization, iGC maintains SHM sequence structure. It also achieves a strong exploration-exploitation balance, yielding competitive reward-diversity products across candidate-selection cutoffs. These results demonstrate that iGC integrates biologically plausible mutation generation with antigen-specific affinity selection, providing an in silico framework for iterative antibody affinity maturation. ## Ablating Design Choices for Protein Ensemble Generators Across Structural Regimes DevFeed: [Ablating Design Choices for Protein Ensemble Generators Across Structural Regimes]() Original publisher: [Read original article]() Author: Eoin Quinn, Oliver Bent Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [model architecture]() Tags: [alignment](), [architecture](), [cs-lg](), [data](), [design](), [domain](), [experimental](), [experts](), [flow](), [flow-matching](), [fragment](), [generative](), [generative-models](), [generators](), [language](), [language-model](), [proteins](), [q-bio-bm](), [sequence](), [shift](), [simulations](), [training]() ### AI overview Across ordered, disordered, multi-domain and transitional proteins, the study finds that training data choices affect ensemble generator performance more than several tested architectural changes. In a minimal Cα flow-matching model evaluated against experimental SAXS and NMR chemical shifts, mixture-of-experts routing, static pair representations and a disorder probe showed no detectable effect. A smaller protein language model performed worse on ordered proteins, while alignment in the loss reduced SAXS agreement. Training on fragment-assembled rather than coarse-grained disordered ensembles improved chemical-shift agreement at a cost to SAXS; an equal mixture retained most of that gain with a smaller SAXS cost. ### Source excerpt arXiv:2611.05055v1 Announce Type: cross Abstract: Generative models of protein conformational ensembles inherit much of their design from structure prediction for folded proteins. Ordered and disordered proteins, however, occupy distinct structural regimes, and we investigate how design choices hold up across them. Using a minimal C$\alpha$ flow-matching model, we ablate choices of architecture, sequence conditioning, rigid-body alignment and training source, and evaluate each variant against experimental SAXS and NMR chemical-shift data partitioned into ordered, disordered, multi-domain and transitional proteins. Mixture-of-experts routing, a static pair representation and a disorder probe have no detectable effect, a smaller protein language model performs worse on ordered proteins but not on disordered ones, and alignment in the loss degrades agreement with SAXS. Replacing coarse-grained simulations of disordered proteins with fragment-assembled ensembles improves agreement with chemical shifts at a cost to SAXS, while an equal mixture of the two keeps most of the gain at a smaller cost. This points to the training data, rather than the architecture, as the main constraint. ## El Agente F\\'armaco: A Scalable AI Scientist for Drug Discovery DevFeed: [El Agente F\\'armaco: A Scalable AI Scientist for Drug Discovery]() Original publisher: [Read original article]() Author: Marcel M\\"uller, Mohammad Ghazi Vakili, Yoshihiro Ogura, Ella Ashoori, Shimeng Wang, Alessandro S. Nascimento, Eddie Calleja, Yutong Zhao, Yimeng Liu, Ella Miray Rajaonson, Aiwei Yin, Jiaru Bai, Yunhe Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [AI for science](), [Multi-step AI processes](), [Agent Harness](), [Large Language Model](), [AI Context Management](), [Retrieval Augmented Generation (RAG)]() Tags: [ai](), [cs-ai](), [cs-ma](), [drug-discovery](), [q-bio-bm]() ### AI overview El Agente Fármaco turns natural-language discovery goals into computational screening workflows: it selects tools, reviews results, and adapts next steps to the target and user objectives. Its shared Fármaco Atlas records molecules, structures, results, and provenance; targeted queries retrieve decision-relevant evidence, while one request can run calculations across a molecule collection without a separate LLM decision for each compound. Integrated tools generate molecules, predict binding and molecular properties, and search for purchasable compounds. In a retrospective comparison, it proposed chemically distinct candidates with competitive docking scores and higher calculated drug-likeness under specified selection criteria. In a 22-task benchmark against plain tool calls and Codex, it had the highest observed mean model-judged score, with lower median cumulative token use and peak context. ### Source excerpt arXiv:2611.03832v1 Announce Type: new Abstract: Computational drug discovery offers a growing range of tools for proposing and evaluating molecules, but combining them and deciding what to do next still requires substantial expert effort. Here, we introduce El Agente F\'armaco, an AI scientist that uses modern large language model (LLM) technology to turn discovery goals expressed in everyday language into complex, scalable computational screening workflows. The agentic harness selects scientific tools, reviews their results, and adjusts subsequent steps to the target and the user's objectives. Our approach centers on a shared scientific memory: F\'armaco Atlas, which stores molecules, structures, and results together with their provenance -- a record of how they were obtained. The agent uses targeted database queries to retrieve the evidence needed for each decision. A single request can launch calculations for an entire collection of molecules, without requiring a separate LLM decision for every compound. This design limits the information the model must process as screening collections grow, while keeping the underlying results available for inspection and further analysis. The integrated tools support molecule generation, prediction of binding and molecular properties, and searches for compounds available for purchase. In a retrospective comparison with an expert-led campaign, the agent proposed chemically distinct candidates with competitive docking scores and higher calculated drug-likeness under explicitly imposed selection criteria. In a 22-task benchmark against plain tool calls and Codex, the agent achieved the highest observed mean model-judged score, with lower median cumulative token use and peak context. F\'armaco provides a foundation for flexible, scalable drug discovery workflows that connect scientific reasoning with an enduring, traceable record of computational work. ## Empirical Bayes Flow Matching for Continuous Cryo-EM Heterogeneity DevFeed: [Empirical Bayes Flow Matching for Continuous Cryo-EM Heterogeneity]() Original publisher: [Read original article]() Author: Daniel Aibinder, Azmi Haider, Dan Rosenbaum Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [VAE](), [Neural Network]() Tags: [3d](), [approximation](), [bayes](), [continuous](), [cs-lg](), [electron](), [expectation](), [flow-matching](), [generation](), [generative](), [images](), [learning](), [linear](), [microscopy](), [model](), [models](), [monte-carlo](), [operator](), [pretraining](), [problems](), [projection](), [q-bio-bm]() ### AI overview EB-FM learns a generative prior for continuous cryo-EM heterogeneity directly from noisy projection images, without pretraining on clean structures. It alternates approximate posterior sampling with guided flow generation and persistent latent estimates (E-step), then updates the flow model using those estimates as training targets (M-step). The authors report validation on two-moons and MNIST inverse problems and simulated cryo-EM data with known poses, where the method recovers continuous protein-conformation distributions in atomic-coordinate space. The supplied abstract does not establish performance on experimental cryo-EM data. ### Source excerpt arXiv:2611.05202v1 Announce Type: new Abstract: We address the problem of learning generative priors over high-dimensional latent variables from indirect, noisy observations, with a focus on continuous heterogeneity in cryo-electron microscopy (cryo-EM). In this setting, 3D atomic structures $x \in \mathbb{R}^{3N_a}$ are never directly observed and must be inferred from 2D projection images $y$ generated by a complex, non-linear forward operator. We propose Empirical Bayes Flow Matching (EB-FM), a method that learns a flow matching prior $p_\theta(x)$ directly from observations using a Monte Carlo stochastic approximation Expectation-Maximization framework. EB-FM alternates between an E-step that performs approximate posterior sampling from $p_\theta(x \mid y)$ via guided flow-based generation while maintaining persistent latent estimates through stochastic approximation, and an M-step that updates the flow model using these stabilized latent variables as training targets. This approach eliminates the need for pretraining on clean data and naturally accommodates non-linear forward models. We validate our method on controlled inverse problems using two-moons and MNIST data, and demonstrate its effectiveness on simulated cryo-EM data with known poses, showing that EB-FM can recover continuous distributions of protein conformations directly in atomic-coordinate space. ## Heterogeneous Graph Attention Network With Multimodal Feature Selection for Cancer Multiomics Integration DevFeed: [Heterogeneous Graph Attention Network With Multimodal Feature Selection for Cancer Multiomics Integration]() Original publisher: [Read original article]() Author: Sina Tabakhi, Charlotte Vandermeulen, Ian Sudbery, Haiping Lu Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Feature Engineering](), [Graphs]() Tags: [attention](), [biomarkers](), [cancer](), [cs-lg](), [cs-ma](), [deep-learning](), [diagnosis](), [feature](), [feature-selection](), [graph](), [integration](), [interpretability](), [machine-learning](), [multimodal](), [network](), [q-bio-bm](), [q-bio-gn]() ### AI overview HeteroGATomics combines multimodal feature selection with heterogeneous graph attention for cancer diagnosis. A multi-agent system builds feature- and patient-similarity networks for each omics modality, then combines them into one heterogeneous graph to learn omics-specific representations and integrate predictions. The abstract reports better results than baseline methods on most metrics across three cancer multiomics datasets, and better results than graph-based integration methods across metrics on an independent breast cancer cohort. It also reports identifying biomarkers that contribute to diagnosis outcomes. The supplied text does not provide dataset details, metric values, or implementation specifics. ### Source excerpt arXiv:2408.02845v2 Announce Type: replace Abstract: The increase in high-dimensional multiomics data demands advanced integration models to capture the complexity of human diseases. Graph-based deep learning integration models, despite their promise, struggle with small patient cohorts and high-dimensional features, often applying independent feature selection without modeling relationships among omics. Furthermore, conventional graph-based omics models focus on homogeneous graphs, lacking multiple types of nodes and edges to capture diverse structures. We introduce a Heterogeneous Graph ATtention network for omics integration (HeteroGATomics) to improve cancer diagnosis. HeteroGATomics performs multimodal feature selection through a multi-agent system, creating dedicated networks of feature and patient similarity for each omics modality. These networks are then combined into one heterogeneous graph for learning holistic omics-specific representations and integrating predictions across modalities. Experiments on three cancer multiomics datasets show that HeteroGATomics outperforms baseline methods across most evaluation metrics in cancer diagnosis. External validation on an independent breast cancer cohort further demonstrates its generalizability, where HeteroGATomics outperforms graph-based multiomics integration methods across evaluation metrics. Moreover, HeteroGATomics enhances interpretability by identifying important biomarkers contributing to the diagnosis outcomes. ## Predicting protein structures at any folding stage from sequence DevFeed: [Predicting protein structures at any folding stage from sequence]() Original publisher: [Read original article]() Author: Lisheng Zhang, Qingsong Li, Zinan Xie, Yanting Yang, Qikui Yang, Peng Lu, Zhican Fu, Wentao Cui, Runtong Hou, Zhengwei Xie Published: 2026-11-04T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Machine learning](), [Artificial Intelligence](), [alphafold2]() Tags: [3](), [about](), [abstract](), [all](), [alpha](), [alphafold](), [alphafold-3](), [also](), [arxiv](), [assembly](), [atom](), [away](), [better](), [binning](), [but](), [c](), [coherent](), [cs-ai](), [folding](), [q-bio-bm](), [sequence](), [stage]() ### AI overview FoldScape generates protein Cα conformational ensembles from sequence at a specified folding fraction, covering stages from collapsed coil to near-native globule. It predicts inter-residue distances in sliding windows, assembles chains, and samples structures with a rectified-flow decoder; training uses coarse-grained folding and all-atom mechanical-unfolding trajectories. On 20 human proteins, it covered about an order of magnitude more of the folding coordinate than seed-varied AlphaFold 3 ensembles at matched sample size, with roughly constant reconstruction accuracy across folding stages. AlphaFold 3 better anchors the immediate native vicinity, which FoldScape does not reach, so the methods are complementary. A stride-one crop-and-superpose scheme extends inference to longer chains. The abstract also reports residue-level mechanical-unfolding resolution and coherent partially unfolded segments; these findings are based on the described trajectories and evaluations. ### Source excerpt arXiv:2611.04287v1 Announce Type: cross Abstract: Protein structure predictors reliably identify native conformations, but do not describe the continuum connecting unfolded ensembles to folded states. Here we present FoldScape, a sequence-conditioned deep learning framework that generates C$\alpha$ conformational ensembles at a specified folding fraction $Q$. FoldScape reconstructs three-dimensional chains through hierarchical, sliding-window regressions of one-dimensional inter-residue distances, followed by chain assembly and sampling with a rectified-flow decoder. It is trained on multi-scale coarse-grained folding trajectories and all-atom mechanical-unfolding trajectories. Across 20 human proteins (36-857 residues), at matched sample size, FoldScape covers about an order of magnitude more of the folding coordinate than seed-varied AlphaFold 3 ensembles, robustly across proteins, order parameters and binning choices. AlphaFold 3 better anchors the immediate native vicinity, which FoldScape does not reach, making the approaches complementary. FoldScape reconstruction accuracy remains approximately constant along $Q$, from collapsed coil to near-native globule, rather than deteriorating away from the native state. A stride-one crop-and-superpose scheme extends inference to chains several times longer than a single chain-level forward pass. We also show that training trajectories resolve mechanical unfolding at residue resolution, that elongation-conditioned ensembles exhibit reproducible continuous observables across pulling velocities, and that partially unfolded chains form geometrically coherent segments. FoldScape thus combines broad coverage of folding stages with stable reconstruction accuracy, without per-protein simulation. ## PeptSol: A Translational Benchmark for Protected-Peptide Solubility in Drug Manufacturing DevFeed: [PeptSol: A Translational Benchmark for Protected-Peptide Solubility in Drug Manufacturing]() Original publisher: [Read original article]() Author: Alexander Arjun Lobo, Marc Saouda, Urvi Awasthi, Leonid Zhukov Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [benchmark overfitting machine learning](), [Representation learning](), [AI Foundation Models]() Tags: [benchmark](), [benchmarks](), [chemistry](), [cs-lg](), [data](), [dataset](), [evaluation](), [machine-learning](), [manufacturing](), [model](), [physics-chem-ph](), [q-bio-bm](), [synthesis]() ### AI overview PeptSol provides an open benchmark for predicting protected-peptide solubility relevant to liquid-phase peptide synthesis, where intermediates must dissolve for reaction and be recoverable for purification. It contains 1,388 curated measurements but covers just 84 parent sequences and 101 protected structures; 92.5% of measurements come from one historical programme. Across seven frozen molecular encoders, the best reported macro-F1 on the historical subset with peptide-held-out splits was 0.843 for binary solubility and 0.614 for the harder three-class ordinal task. No foundation model consistently beat a baseline combining chemical fingerprints and explicit molecular descriptors, and model rankings changed across tasks and splits. The benchmark, protocol, and baselines offer a way to measure progress while these coverage limits constrain how broadly results can be generalized. ### Source excerpt arXiv:2611.00749v1 Announce Type: new Abstract: Liquid-phase peptide synthesis (LPPS) offers a scalable route to peptide manufacturing, but route development still depends on empirical solvent screens because each protected intermediate must be soluble for reaction and recoverable for purification. The protecting groups, terminal modifications, and noncanonical residues that govern this behaviour are poorly represented in existing peptide property benchmarks, limiting the ability to develop and rigorously evaluate molecular representation models for LPPS-relevant prediction tasks. We introduce PeptSol, an open benchmark of 1388 hand-curated protected-peptide solubility measurements with chemistry-aware splits. The measurements span only 84 parent sequences and 101 protected structures, and 92.5% come from a single historical programme. Across seven frozen molecular encoders, the best macro-F1 is 0.843 on binary solubility for the historical subset under a peptide-held-out split; the harder three-class ordinal task reaches 0.614 in that cell. No foundation model consistently outperforms a baseline combining chemical fingerprints with explicit molecular descriptors, and model rankings vary across prediction tasks and data splits. The dataset, evaluation protocol, and baselines establish a standard for measuring progress on molecular representations for synthesis-aware peptide developability. ## D2ML: Models for Direct-to-Biology Small Molecule Screening DevFeed: [D2ML: Models for Direct-to-Biology Small Molecule Screening]() Original publisher: [Read original article]() Author: AkshatKumar Nigam, Neri Amara, Hillary Dequina, Dyana Kenanova, Seth Malmersjo, Fiorella Ruggiu, Raymond V. Fucini, Jack Sadowsky, Stig K. Hansen, Imran S. Haque Published: 2026-11-02T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Neural Network]() Tags: [biology](), [cs-lg](), [experiment](), [machine-learning](), [models](), [neural-networks](), [optimization](), [prediction](), [q-bio-bm](), [training-data]() ### AI overview Assay-aware models use different stages of direct-to-biology screening to predict activity from single-point screens, relative potency across targets in crude dose-response assays, and absolute potency from purified compounds. Modeling potency ratios for targets tested from the same crude well can preserve selectivity information despite uncertain concentrations. A joint model sharing information across assay stages improved predictions over separate neural networks; crude dose-response selectivity rank correlation rose from 0.27 to 0.40. In a separate test, models guided selection of 84 compounds targeting ROS1 potency while limiting TRKB activity. The most selective new compound reached a ROS1 IC50 of 14.7 nM and 233-fold selectivity, compared with 20.3 nM and 79-fold for the most selective training compound. The abstract describes ATLAS as supporting cycles that synthesize and screen 1,000-10,000 selected compounds in two to three weeks. ### Source excerpt arXiv:2610.41070v1 Announce Type: cross Abstract: Direct-to-biology (D2B) screening enables rapid testing of newly synthesized molecules without purification. However, uncertain concentrations in crude mixtures and limited purified-compound measurements complicate the use of these data in machine learning. Here we show that models tailored to the structure of D2B data can improve prediction and guide the discovery of more potent and selective compounds. Our assay-aware models predict activity from single-point screens, relative potency across targets from crude dose-response assays, and absolute potency from purified-compound measurements. For crude assays, potency ratios between targets tested from the same well can retain selectivity information despite uncertain concentrations. A joint model that shares information across stages and balances their contributions to training improved predictions over separate neural networks. The largest gain was in crude dose-response selectivity prediction, where rank correlation increased from 0.27 to 0.40. In a separate experiment, independently trained models guided the selection of 84 new compounds to improve ROS1 potency while limiting activity against the anti-target TRKB. Synthesis and testing identified compounds that extended the measured Pareto front, achieving better combinations of potency and selectivity than those in the training data. The most selective new compound achieved a ROS1 IC50 of 14.7 nM and 233-fold selectivity over TRKB. By comparison, the most selective training compound had a ROS1 IC50 of 20.3 nM and 79-fold selectivity. ATLAS provides the experimental setting for this approach, supporting synthesis and screening of 1,000-10,000 model-selected compounds in two- to three-week cycles. Together, these results show how assay-aware learning can use the full D2B assay sequence to guide multi-objective molecular optimization. ## When Does Context Help? A Systematic Study of Target-Conditional Molecular Property Prediction DevFeed: [When Does Context Help? A Systematic Study of Target-Conditional Molecular Property Prediction]() Original publisher: [Read original article]() Author: Bryan Cheng, Jasper Zhang Published: 2026-11-02T05:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Estimator](), [ai evaluation](), [few-shot]() Tags: [2020](), [2021](), [absolute](), [abstract](), [architecture](), [architectures](), [arxiv](), [based](), [benchmarking](), [can](), [concatenation](), [conditional](), [context](), [cs-lg](), [data](), [distribution](), [help](), [prediction](), [property](), [q-bio-bm](), [q-bio-mn](), [study](), [target]() ### AI overview Across 10 protein families and data regimes of 67-9,409 compounds, the study finds that how target context is incorporated matters more than simply including it: NestDrug's FiLM fusion outperforms concatenation by 24.2 percentage points and additive conditioning by 8.6 points. Context can help when per-target data are scarce: for CYP3A4, multi-task transfer reaches 0.686 AUC versus 0.238 for a per-target Random Forest. It can also hurt: distribution mismatch degrades BACE1 performance by 10.2 points, and few-shot adaptation underperforms zero-shot. The authors flag benchmark leakage and an unusually strong non-learning baseline: 1-nearest-neighbor Tanimoto reaches 0.991 AUC on DUD-E, with 50% of actives leaking into training. On a temporal split (training through 2020, testing 2021-2024), performance is 0.843 AUC without degradation. These findings are based on the supplied abstract. ### Source excerpt arXiv:2604.06558v2 Announce Type: replace Abstract: We present the first systematic study of when target context helps molecular property prediction, evaluating context conditioning across 10 diverse protein families, 4 fusion architectures, data regimes spanning 67-9,409 training compounds, and both temporal and random evaluation splits. Using NestDrug, a FiLM-based architecture that conditions molecular representations on target identity, we characterize both success and failure modes with three principal findings. First, fusion architecture dominates: FiLM outperforms concatenation by 24.2 percentage points and additive conditioning by 8.6 pp; how you incorporate context matters more than whether you include it. Second, context enables otherwise impossible predictions: on data-scarce CYP3A4 (67 training compounds), multi-task transfer achieves 0.686 AUC where per-target Random Forest collapses to 0.238. Third, context can systematically hurt: distribution mismatch causes 10.2 pp degradation on BACE1; few-shot adaptation consistently underperforms zero-shot. Beyond methodology, we expose fundamental flaws in standard benchmarking: 1-nearest-neighbor Tanimoto achieves 0.991 AUC on DUD-E without any learning, and 50% of actives leak from training data, rendering absolute performance metrics meaningless. Our temporal split evaluation (train up to 2020, test 2021-2024) achieves stable 0.843 AUC with no degradation, providing the first rigorous evidence that context-conditional molecular representations generalize to future chemical space. ## What to randomize matters: a controlled study of volume, axis and representation in low-data molecular generation DevFeed: [What to randomize matters: a controlled study of volume, axis and representation in low-data molecular generation]() Original publisher: [Read original article]() Author: Neerav Kaushal, Alisa Teige, Ajay MNV Penmatsa Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [data augmentation](), [machine learning overfitting]() Tags: [autoregressive](), [cs-ce](), [cs-cl](), [cs-lg](), [data](), [generation](), [machine-learning](), [q-bio-bm](), [q-bio-qm]() ### AI overview Using an autoregressive sequence model trained on 1,000 seed molecules, the study compares SAFE representations and randomized SMILES for low-data molecular generation. Randomizing SAFE increased the genuine yield of novel molecules by more than 450-fold over canonical SAFE. Atom-order and fragment-order randomization worked jointly: each axis alone achieved only a fraction of the yield when both varied. With matched training corpora and compute, randomized SAFE produced 2.8 times the genuine yield of randomized SMILES. These findings concern the study's reported low-data setup; the supplied text does not describe broader validation. ### Source excerpt arXiv:2610.38835v1 Announce Type: new Abstract: Chemical language models can struggle to generate new molecules when trained on small datasets, where a model may reproduce training examples rather than learn a useful generative distribution. Randomized SMILES augmentation varies atom traversal, whereas Sequential Attachment-based Fragment Embedding (SAFE) also allows the order of fragment blocks to vary. How these distinct sources of representational variation affect low-data molecular generation has not been clearly studied. Here, using 1,000 seed molecules to train an autoregressive sequence model, we compare canonical SAFE with randomized SAFE at increasing corpus sizes, isolate atom-order and fragment-order axes of SAFE randomization, and compare fully randomized SAFE with randomized SMILES at molecule-by-molecule matched training corpora. Three findings emerge. First, randomization of SAFE alone raises the genuine yield of novel molecules more than 450-fold, turning a model that merely reproduces its training set into a practical molecular generator. Second, the two axes of randomization act jointly rather than additively: atom order alone reaches a tenth, and fragment order alone a seventh, of the genuine yield obtained when both vary, which is four times their combined contribution. Third, at matched corpus size and compute, randomized SAFE delivers 2.8 times the genuine yield of randomized SMILES. Welch tests with Games-Howell post-hoc comparisons resolve every difference on the primary outcomes. In the low-data regime, thus, what one randomizes rather than only how much determines whether augmentation produces new molecules or merely new spellings of the training set. ## Bridging the Simulation-to-Experiment Gap with Generative Models using Adversarial Distribution Alignment DevFeed: [Bridging the Simulation-to-Experiment Gap with Generative Models using Adversarial Distribution Alignment]() Original publisher: [Read original article]() Author: Kai Nelson, Tobias Kreiman, Sergey Levine, Aditi S. Krishnapriyan Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [AI Models](), [Generative Adversarial Network](), [Simulation]() Tags: [adversarial](), [alignment](), [approach](), [atomic](), [bridging](), [challenge](), [cond-mat-mtrl-sci](), [cs-lg](), [data](), [data-driven](), [distribution](), [domain](), [engineering](), [experiment](), [experimental](), [framework](), [gap](), [generative](), [generative-models](), [measurements](), [models](), [q-bio-bm](), [simulation]() ### AI overview Adversarial Distribution Alignment (ADA) addresses the simulation-to-experiment gap by pretraining a generative model on fully observed but imperfect simulation data, then aligning its distribution with partial observations from real experiments. In the physical-sciences application, the model starts from atomic positions sampled from a simulated Boltzmann distribution. The authors prove recovery of the target observable distribution, including when observables are multiple and potentially correlated, and report empirical validation on synthetic, molecular, and experimental protein data. The excerpt does not provide implementation details or quantify performance. ### Source excerpt arXiv:2604.01169v2 Announce Type: replace Abstract: A fundamental challenge in science and engineering is the simulation-to-experiment gap. While we often possess prior knowledge of physical laws, these physical laws can be too difficult to solve exactly for complex systems. Such systems are commonly modeled using simulators, which impose computational approximations. Meanwhile, experimental measurements more faithfully represent the real world, but experimental data typically consists of observations that only partially reflect the system's full underlying state. We propose a data-driven distribution alignment framework that bridges this simulation-to-experiment gap by pre-training a generative model on fully observed (but imperfect) simulation data, then aligning it with partial (but real) observations of experimental data. While our method is domain-agnostic, we ground our approach in the physical sciences by introducing Adversarial Distribution Alignment (ADA). This method aligns a generative model of atomic positions -- initially trained on a simulated Boltzmann distribution -- with the distribution of experimental observations. We prove that our method recovers the target observable distribution, even with multiple, potentially correlated observables. We also empirically validate our framework on synthetic, molecular, and experimental protein data, demonstrating that it can align generative models with diverse observables. Our code is available at https://kaityrusnelson.com/ada/. ## Latent Style-based Quantum Wasserstein GAN for Drug Design DevFeed: [Latent Style-based Quantum Wasserstein GAN for Drug Design]() Original publisher: [Read original article]() Author: Julien Baglio, Yacine Haddad, Richard Polifka Published: 2026-10-30T04:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Quantum Computing](), [Artificial Intelligence](), [VAE]() Tags: [adversarial](), [artificial-intelligence](), [autoencoder](), [capacity](), [circuits](), [cs-ai](), [cs-lg](), [design](), [drug-discovery](), [drugs](), [gans](), [heron](), [ibm](), [inference](), [metrics](), [noise](), [performance](), [pipeline](), [q-bio-bm](), [quant-ph](), [quantum](), [qubits](), [sampling](), [training]() ### AI overview The proposed latent style-based quantum Wasserstein GAN combines a variational autoencoder for mapping molecules into latent space with a hybrid quantum GAN. It reuploads style-based noise at each rotational gate and uses a Wasserstein gradient penalty to improve trainability and mitigate mode collapse. On noiseless simulators, the authors trained and sampled with circuits up to 15 qubits; an inference-only IBM Heron run used 5 qubits to sample 2,500 molecules and examine quantum noise. The 110-parameter generator achieved statistically comparable overall performance and significant gains on selected drug-relevant metrics versus both the roughly 700,000-parameter MOSES latent GAN and a classical latent GAN matched for generator capacity. The excerpt reports these as design-model results, not evidence of clinical efficacy. ### Source excerpt arXiv:2603.22399v2 Announce Type: replace-cross Abstract: The development of new drugs is a time-consuming and expensive process, with average costs estimated to be up to around $2.5 billion. The first step in this process is the design of the new drug, for which de novo drug design, assisted by artificial intelligence, has blossomed in recent years and has revolutionized the field. In particular, generative artificial intelligence has delivered promising results in drug discovery and development, reducing costs and time to solution. However, classical generative models, such as generative adversarial networks (GANs), can be difficult to train and are prone to mode collapse. Quantum computing may be an avenue to overcome some of these issues, providing compact models with far fewer trainable parameters, thereby enhancing the generalizability of GANs. We propose a new latent style-based quantum Wasserstein GAN (QGAN) for drug design, combining a variational autoencoder that maps molecular structures into a latent space with a hybrid quantum GAN. Our pipeline is benchmarked against classical models of the industry-relevant MOSES suite. Style-based noise reuploading at every rotational gate and a Wasserstein gradient penalty are used to improve trainability and to mitigate mode collapse. Training and large-scale sampling with circuits of up to 15 qubits are performed on noiseless quantum simulators to validate our architecture. We present an inference-only demonstration on an IBM Heron quantum computer using 5 qubits, sampling 2,500 new molecules, to investigate quantum noise effects. Our default QGAN, using 110 trainable generator parameters, reaches statistically comparable overall performance and statistically significant gains on selected drug-relevant metrics relative to both the MOSES classical latent GAN baseline, which contains about 700,000 trainable parameters for the generator, and a classical latent GAN with matched generator capacity. ## What Transfers Across Protein Pockets? Experience Reuse in Molecular Optimization for Agentic Structure-Based Drug Design DevFeed: [What Transfers Across Protein Pockets? Experience Reuse in Molecular Optimization for Agentic Structure-Based Drug Design]() Original publisher: [Read original article]() Author: Zaifei Yang, Weiyu Chen, Yaqing Wang, James Kwok Published: 2026-10-26T04:00:00Z Content type: news Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [llm-reasoning](), [Loop Engineering]() Tags: [agent](), [agentic](), [agents](), [artificial-intelligence](), [cs-ai](), [design](), [experience](), [llm](), [llm-agents](), [optimization](), [q-bio-bm](), [reuse](), [structure]() ### AI overview The study finds that experience reuse across protein pockets may help rank an agent's proposed ligand edits for docking more than dictate what to build. An analysis of a run across CrossDocked2020 test-set pockets found that keys combining site contacts, edit action, and round sharply separated improving from worsening edits, but this was largely explained by edit size, action, and round rather than target properties. Naming the target separated outcomes several times more sharply, though that identity is unavailable for an unseen pocket. Of the exact-match descriptions tested, specific ones rarely recurred and coarse ones separated outcomes weakly; transferable signals mainly described the edit and resulting molecule. These conclusions are limited to the representations and lookup scheme studied. ### Source excerpt arXiv:2610.32620v1 Announce Type: new Abstract: LLM agents optimize a ligand in a protein pocket by editing it step by step and docking each edit. Recent agents first spend extra dockings probing the complex with controlled edits; guidance induced from these probes is reported to help more than the chemical knowledge the model already carries. The probes are then discarded with the instance that produced them, so the obvious next step is to keep them in an experience base, letting a new pocket inherit what earlier ones already paid for. Such a base is read through a \emph{key}: the record properties under which past edits are pooled and a new one is looked up. Using the full log of one such run over the CrossDocked2020 test-set pockets, we ask whether its keys recur on a pocket the base has never seen and still separate improving from worsening edits. \crchange{Using} the contacts at the site being edited, the action, and the round as \crchange{a} key, the base separates them sharply. Within our analysis, this separation is largely explained by how much the edit changed the molecule's size, the action, and the round, none of which describes the target. That is not because the target is irrelevant: naming it outright separates them several times more sharply. For a pocket absent from the base, transfer must rely on a description rather than its identity. Among the exact-match descriptions we evaluate, the more specific ones rarely recur, whereas the coarser ones provide little separation; the signals that do transfer primarily describe the edit and the molecule it produced rather than the target. Under the representations and lookup scheme studied here, an experience base may be more useful for prioritizing an agent's own proposals for docking than for directly prescribing what to build. ## Probe Before You Edit: Probing-Guided Molecular Optimization for LLM Agents in Structure-Based Drug Design DevFeed: [Probe Before You Edit: Probing-Guided Molecular Optimization for LLM Agents in Structure-Based Drug Design]() Original publisher: [Read original article]() Author: Zaifei Yang, Weiyu Chen, Yaqing Wang, James Kwok Published: 2026-10-26T04:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Multi Agent Systems]() Tags: [affinity](), [agent](), [agents](), [analog](), [benchmark](), [complex](), [cs-ai](), [design](), [diagnostics](), [failure](), [framework](), [improvement](), [liability](), [llm](), [llm-agents](), [local](), [map](), [metrics](), [multi-agent](), [objectives](), [optimization](), [q-bio-bm]() ### AI overview PROBE addresses the difficulty of improving binding affinity and druggability together during ligand optimization. It maps editable ligand sites to identify plausible joint gains, likely tradeoffs, and liability substructures, then makes controlled probe edits and distills their responses into an EditManual. An iterative multi-agent optimization loop uses both guides. The abstract reports state-of-the-art performance on CrossDocked2020 and says PROBE substantially mitigates the failure modes measured by its diagnostic metrics. ### Source excerpt arXiv:2606.00555v3 Announce Type: replace Abstract: Structure-based drug design increasingly employs LLM agents to iteratively refine ligands against a target pocket, yet a viable ligand must satisfy two often-conflicting objectives---binding affinity and druggability---which single optimization steps rarely improve together. To quantify this difficulty, we introduce two diagnostic metrics: the first measures how often a single edit improves both objectives, and the second measures how often a gain on one objective comes with a loss on the other. Applying these diagnostics to current LLM-agent pipelines exposes a consistent failure mode: the agent performs molecular editing without knowing how the pocket-ligand complex responds to local modifications, thus rarely achieving joint improvement. Inspired by medicinal chemists, who probe the pocket-ligand complex with controlled analog edits before choosing an optimization direction, we propose \textbf{PROBE}, an optimization framework built around edit--response probing. PROBE first decomposes the ligand into editable sites and builds a pocket-specific \textbf{site map} that flags where joint gains are plausible, where the two objectives are likely in tension, and where liability substructures should be changed; it then performs controlled probe edits whose responses are distilled into an \textbf{EditManual}. Guided by the site map and EditManual, PROBE runs an iterative multi-agent optimization loop. On the CrossDocked2020 benchmark, PROBE achieves state-of-the-art performance and substantially mitigates the failure modes exposed by our diagnostics metrics. Codes are in https://github.com/yzf-code/PROBE/tree/main ## Advances in Machine Learning for Directed Evolution: A Five-Year Retrospective DevFeed: [Advances in Machine Learning for Directed Evolution: A Five-Year Retrospective]() Original publisher: [Read original article]() Author: Bruce J. Wittmann Published: 2026-10-23T04:00:00Z Content type: opinion Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Model Development]() Tags: [cost](), [cs-lg](), [dna](), [engineering](), [evolution](), [goals](), [machine-learning](), [ml](), [objectives](), [practical](), [q-bio-bm](), [resource](), [retrospective](), [sufficient](), [synthesis](), [time]() ### AI overview The retrospective argues that machine learning has had less impact on directed evolution partly because ML-assisted directed evolution (MLDE) often seeks an optimal protein, while practical directed evolution seeks a sufficient protein within time and resource constraints. It highlights DNA synthesis costs as a key omission in nearly all current MLDE methods, limiting practical applicability regardless of model capability. The author notes recent exceptions and argues that the past five years of ML-assisted protein engineering can coexist with reframing MLDE objectives around practical constraints. ### Source excerpt arXiv:2609.03046v2 Announce Type: replace-cross Abstract: The last five-plus years have seen many protein engineering disciplines transformed by advances in machine learning (ML), but the same cannot be said for directed evolution. Reflecting on a previously co-authored perspective, I discuss why I believe this to be the case, arguing that a disconnect between the goals of machine-learning-assisted directed evolution (MLDE) researchers--"identify an optimal protein"--and the goals of directed evolution more broadly--"identify a sufficient protein given time and resource constraints"--is a principal culprit. As an example, I highlight how nearly all current MLDE methods neglect to account for the cost of DNA synthesis, resulting in strategies that have limited practical applicability regardless of the underlying models' capabilities. I close by discussing recent works that are exceptions to this overarching trend, and emphasize that the last five years of efforts in ML-assisted protein engineering and the prescribed reframe of MLDE objectives need not be mutually exclusive. ## AI Mindmeld: A Latent Interconnect Protocol DevFeed: [AI Mindmeld: A Latent Interconnect Protocol]() Original publisher: [Read original article]() Author: Siavash Golkar, Jake Kovalic, Shirley Ho Published: 2026-10-23T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Representation learning](), [Large Language Model]() Tags: [agents](), [ai](), [connectors](), [cs-lg](), [exchange](), [gemma-3](), [interconnect](), [kimi-k2-6](), [llms](), [protocol](), [q-bio-bm](), [specialists](), [stat-ml](), [zero-shot]() ### AI overview The Latent Interconnect Protocol (LIP) maps specialist model representations into an LLM's internal states through independently trained connectors, enabling direct exchange of information without joint training or changes to pretrained models. The authors connect protein and small-molecule specialists to Gemma 3 and Kimi K2.6, reporting useful responses to familiar and new questions about unseen objects and zero-shot transfer to cross-domain tasks. Performance and compositionality generally improve with LLM capability, especially for zero-shot transfer. A key limitation is that current LLMs were not trained to accept specialist representations as native inputs, so access does not ensure they can interpret or use them reliably. The authors call for consistent specialist interfaces and LLM training to receive, combine and reason with these representations. ### Source excerpt arXiv:2610.30215v1 Announce Type: new Abstract: LLMs can act as scientific agents using tools to consult specialists and run experiments. Yet specialist information must enter their context, generally as text, images, audio or video: a bottleneck between models with rich, high-dimensional representations. We advocate direct representation exchange: a mindmeld conveying many properties, including information an LLM would not think to request. It should be extensible to new specialists, flexible across questions and tasks, and compositional across independently connected specialists without joint training. The Latent Interconnect Protocol (LIP) learns connectors mapping specialist representations into LLM internal states. Training only connectors keeps this practical as LLMs grow and evolve. We connect three protein and three small molecule specialists to Gemma 3 (4B and 12B parameters) and Kimi K2.6 (1T total, 32B active), freezing all pretrained models. Independently trained connectors provide useful information without changing the LLM or existing connectors, supporting familiar and new question types about unseen objects. We demonstrate zero-shot transfer to cross-domain tasks without joint training. Task performance and compositionality generally improve with LLM capability, with especially strong gains in zero-shot task transfer. Limitations remain: current LLMs were not trained for specialist representations as native inputs; access does not ensure reliable interpretation, integration or use in reasoning. We call on specialist developers to provide consistent representation interfaces and, especially, on LLM developers to train models to receive, combine and reason with these inputs. These efforts could expand scientific agents' information while preserving conventional tools' flexibility. Data, pretrained connectors, and code for training, evaluation and interactive inference: https://github.com/PolymathicAI/AI-Mindmeld ## RiboBridge: protein-conditioned de novo RNA binder structure design DevFeed: [RiboBridge: protein-conditioned de novo RNA binder structure design]() Original publisher: [Read original article]() Author: Zhou Zhang, Hanqun Cao, Minghao Sun, Odin Zhang, Yang Zhang, Pheng Ann Heng, Wujun Li, Tianfan Fu Published: 2026-10-23T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Electrical engineering and computer science (EECS)]() Tags: [arxiv](), [assembly](), [benchmark](), [bind](), [binder](), [bridge](), [candidate](), [canonical](), [complex](), [component](), [cross](), [cs-lg](), [design](), [designing](), [diagnostics](), [error](), [evaluation](), [generation](), [interaction](), [q-bio-bm](), [rna](), [structure]() ### AI overview RiboBridge separates RNA shape generation from assembly with a target protein. It proposes candidate protein-surface interfaces, generates canonical RNA conformations in the pretrained RiboSphere latent space, predicts interaction constraints across the separate coordinate frames, and uses rigid placement to form a protein-relative pose. On a benchmark of 124 protein-RNA complexes, it achieved the best reported fNAT and DockQ and lowest interface and ligand RMSDs among compared methods under post-hoc interface-correspondence evaluation. In that evaluation, native information selects among candidates only after generation; the reported ranking is therefore specific to this protocol. Ablations and stage-wise diagnostics examine contributions and sources of recovery error. ### Source excerpt arXiv:2610.30668v1 Announce Type: new Abstract: Designing RNA molecules that bind to target protein structures requires generating both an RNA conformation and its bound complex geometry, without assuming a predefined binding site. Existing approaches generally model these coupled processes within a unified representation, entangling RNA conformation generation with protein-relative placement. In this work, we introduce RiboBridge, a structure-first framework that separates RNA conformational generation from protein--RNA assembly through cross-frame interaction geometry. RiboBridge first proposes multiple candidate RNA-binding interfaces on the protein surface, generates canonical RNA conformations within a pretrained RNA structural latent space (RiboSphere), and then predicts protein--RNA interaction constraints to bridge the independently defined RNA and protein coordinate frames. The resulting interaction geometry is converted into a geometrically consistent protein-relative pose through rigid placement. We evaluate RiboBridge on a temporally split, sequence-filtered, and redundancy-controlled benchmark of 124 protein--RNA complexes. Under post-hoc interface-correspondence evaluation, where native information is used only after generation to select among already generated predicted-pocket candidates, RiboBridge achieves the highest fNAT and DockQ and the lowest interface and ligand RMSDs among compared methods. Component ablations and stage-wise diagnostics further characterize the contributions of RNA conformational generation, interaction prediction, and rigid placement, identifying the main sources of complex-recovery error. ## Escaping the Phylogenetic Trap: Persistent Homology Enables Cross-Species Ion Channel Classification via Multimodal Graph Neural Networks DevFeed: [Escaping the Phylogenetic Trap: Persistent Homology Enables Cross-Species Ion Channel Classification via Multimodal Graph Neural Networks]() Original publisher: [Read original article]() Author: Arezoo Bybordi, Favour Achimba, Mariam Gelashvili, Jessy Ramirez, Weigang Qiu, Mand\\"e Holford, Anita Raja Published: 2026-10-23T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Artificial Intelligence](), [Representation learning](), [Neural Network](), [Embeddings](), [graph theory]() Tags: [2](), [3](), [abstract](), [all](), [arxiv](), [attention](), [benchmark](), [best](), [channel](), [classification](), [cross](), [cs-lg](), [graph](), [graph-neural-networks](), [homology](), [ion](), [multimodal](), [networks](), [neural](), [neural-networks](), [persistent-homology](), [q-bio-bm]() ### AI overview Random train/test splits mask poor cross-species generalization in venom compound classification: although all models score 81-86% on i.i.d. splits, a sequence-only baseline falls to about 30% under leave-one-taxon-out evaluation. On 1,373 predicted compounds spanning five taxa, the multimodal PH-GNN combines ESM-2 sequence embeddings with a disulfide-aware graph attention network and persistent-homology descriptors of backbone geometry. It reaches 63.4% under taxonomic shift, versus 54-55% for the other geometric priors, a gain of more than 8 percentage points over the next-best method. The abstract attributes this result to topological descriptors' robustness to local structural perturbations and argues that random-split evaluation can obscure differences in representation quality. ### Source excerpt arXiv:2610.30394v1 Announce Type: new Abstract: Protein classifiers are highly sensitive to evaluation protocol: random train/test splits induce phylogenetic leakage, allowing models to exploit taxon-specific correlations and substantially overestimate out-of-distribution performance. This issue is particularly acute for venom compound classification by ion-channel target (Ca2+, K+, Na+), where reported accuracies of 80-90% do not reflect generalization to unseen taxa. We formulate cross-species prediction as a structured distribution shift in which taxonomic identity acts as a confounder, requiring representations that capture structural features invariant across phylogeny. Using a shared multimodal backbone that fuses ESM-2 sequence embeddings with a disulfide-aware graph attention network, we systematically compare five structural inductive biases (spectral, Riemannian, SE(3)-invariant, statistical-manifold, and persistent-homology representations) under a leave-one-taxon-out evaluation protocol on a benchmark of 1,373 ColabFold-predicted venom compounds spanning five taxa. The persistent-homology representation summarizes backbone geometry through topological descriptors that are robust to local structural perturbations, providing an inductive bias toward phylogenetically invariant learning. Under conventional i.i.d. splits, all models achieve similar accuracy (81-86%), masking differences in inductive bias. Under taxonomic shift, this parity collapses: a sequence-only baseline drops to approximately 30%, the remaining geometric priors cluster at 54-55%, while our persistent-homology model (PH-GNN) achieves 63.4%, outperforming the next-best method by over 8 percentage points. These results demonstrate that random-split evaluation obscures meaningful differences in representation quality and identify topological representations as an effective inductive bias for learning under structured biological distribution shift. ## ProMORNA: An End-to-End Generative Framework for Full-Length mRNA Design DevFeed: [ProMORNA: An End-to-End Generative Framework for Full-Length mRNA Design]() Original publisher: [Read original article]() Author: Zixi Shao, Tao Wang, Ye Shen, Tianyi Huang, Yibei Xiao Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Large Language Model](), [Fine-tuning](), [model architecture](), [Advanced AI workflows]() Tags: [architecture](), [benchmark](), [cs-ai](), [cs-lg](), [design](), [efficiency](), [evaluations](), [experiments](), [fine-tuning](), [framework](), [generative](), [generative-models](), [inference](), [language-model](), [large-language-model](), [learning](), [messenger](), [metric](), [model](), [post-training](), [q-bio-bm](), [reinforcement-learning](), [rna]() ### AI overview ProMORNA generates full-length mRNA in a zero-shot framework using a BART-style model pretrained on over 6 million natural protein-mRNA pairs, then cold-start supervised fine-tuning and multi-objective reinforcement learning. The post-training method aggregates standardized relative advantages per metric with domain-specific stabilization to address conflicting biological objectives without directly scalarizing raw scores. In silico, the authors report a predicted stability-translation efficiency Pareto frontier beyond standard baselines and GEMORNA. Across three independent in vitro Fluc2P experiments, Candidate 2 had 3.33 times the 12-72-hour reporter-activity AUC of the conventional benchmark (Holm-adjusted P=0.01285); its estimated 1.29-fold advantage over GEMORNA's reported best candidate is exploratory. ### Source excerpt arXiv:2605.01513v2 Announce Type: replace Abstract: Designing therapeutic messenger RNA (mRNA) requires generating full-length transcripts that optimally balance intracellular stability and translation efficiency. While generative models have advanced mRNA design, existing approaches rely on region-wise generation, expensive inference-time search, or target-specific model updates. To overcome these limitations, we introduce ProMORNA, a zero-shot, end-to-end generative framework for full-length mRNA design. ProMORNA aligns the design process with modern Large Language Model training pipelines: it utilizes a BART-style architecture pre-trained on over 6 million natural protein-mRNA pairs, followed by a two-stage post-training process comprising cold-start supervised fine-tuning and multi-objective reinforcement learning. By standardizing and aggregating relative advantages at the individual metric level with domain-specific stabilization, our post-training stage directly aligns the generative policy with conflicting biological constraints while avoiding direct scalarization of raw objective scores. In silico evaluations demonstrate that ProMORNA pushes the Pareto frontier of predicted stability and translation efficiency beyond standard baselines and GEMORNA, a leading state-of-the-art method. In three independent in vitro Fluc2P experiments, Candidate 2 achieved a 12-72h reporter-activity AUC 3.33 times that of the conventional Benchmark (Holm-adjusted $P=0.01285$), while its estimated 1.29-fold advantage over the reported best GEMORNA candidate remained exploratory. These results support ProMORNA as a promising framework for generating functionally active full-length mRNA candidates.