Weekly BioML Digest [November 10, 2026]

Machine Learning × Computational Biology paper compilation

Hey! It's your weekly digest of machine learning papers in CompBio and Drug Discovery.

Feedback? Email me at biomldigest@gmail.com.

📚 Peer-Reviewed Journals (Top 20)

1190 matched filters -> 20 selected after LLM relevance + novelty ranking.

• 📡 Effective enzyme engineering through modeling kinetic parameter changes upon mutations with geometric deep learning
  Yuan, Qianmu, Zhu, Mingming, ..., Lu, Yutong, Yang, Yuedong — Nature Communications, 2026-11-05
  Introduces DeltaCata, a Siamese geometric graph neural network that directly predicts mutation-induced changes in enzyme kinetics (Δkcat, ΔKm) from structure, guiding beneficial mutation screening and engineering of a methyltransferase with up to 7× improved catalytic efficiency.

• 🧬 Multiscale multimodal integration of spatial omics through a deep generative model with CORAL
  He, Siyu, Bieniosek, Matthew, ..., Trevino, Alexandro E., Zou, James — Nature Genetics, 2026-11-05
  Presents CORAL, a probabilistic deep generative model with graph attention that integrates multiscale spatial multi-omics (transcriptomics, proteomics, metabolomics, epigenomics) to map tissue domains and cell–cell interactions, revealing immune architectures linked to immunotherapy resistance.

• 🌿 An AI-native experimental laboratory for autonomous biomolecular operations
  Wu, Mingyu, Wang, Zhaoguo, ..., Fan, Chunhai, Chen, Haibo — Nature, 2026-11-04
  AutoDNA is an AI-native autonomous biomolecular lab that co-orchestrates >20 instruments across >11,000 steps using end-to-end models and planning, matching expert DNA synthesis yields and completing a full DNA data storage cycle with minimal human intervention.

• 📰 Constrained diffusion-enabled construction of a highly expressed and robust peptide asparaginyl ligase.
  Shuo Pang, Yuxing Hao, Jiarong Mo, Yingying He, Yinglu Cui, Bian Wu, Xinya Hemu — Chemical science, 2026-11-04
  Ernie-Switcher couples a foundation RNA language model with biophysical constraints to navigate the RNA device landscape, designing toehold switches with up to 35-fold ON/OFF ratios validated in E. coli.
  Affiliations: State Key Laboratory of Natural Medicines, School of Traditional Chinese Pharmacy; State Key Laboratory of Microbial Diversity and Innovative Utilization, Institute of Microbiology; ...

• 📡 Genomic language models decode bidirectional promoter activity and reveal conserved regulatory patterns across plant species
  Gao, Shang, Yu, Tingxi, ..., McLachlan, Geoffrey J., Li, Huihui — Nature Communications, 2026-11-05
  Uses genomic language model embeddings (Evo2) to predict bidirectional promoter activity across plants, transferring across species and validated in Spartina, linking model features to nucleosome positioning and TF landscapes.

• 📰 PABbind: A Hierarchical Geometric Framework for Mutation-Involved Antigen−Antibody Interaction Prediction and SARS-CoV-2 Immune Escape Assessment
  Qi-Rui Deng, Su-Tong Xiang, Jing-Xuan Ge, Ting-Jun Hou, Jin-Ping Lei, Hui-Yong Sun — Journal of Medicinal Chemistry, 2026-11-04
  PABbind is a hierarchical geometric deep-learning model that predicts antigen–antibody binding changes for variant antigens using only wild-type complex structures and mutant sequences, enabling large-scale immune escape assessment for SARS-CoV-2 antibodies.

• 📰 MultiMHC: a multiscale learning framework integrating sequence and residue contact topology for peptide–MHC binding prediction
  Wen-Fan Chen, Ying Chi, Yi-Hao Wang, Jing-Cheng Wu, ..., Xue-Li Bai, Shu-Qing Chen, Zhan Zhou, Ru-Hong Zhou — Briefings in Bioinformatics, 2026-11-06
  MultiMHC integrates sequence features with residue-contact topology via multiscale deep learning to predict peptide–MHC binding and enrich neoantigens, improving ranking by incorporating lightweight structural interaction patterns.

• 📰 Identifying multi-omics spatial domains via masked autoencoder and diffusion-based latent regularization with SpaMEDM
  Chang Zhou, Ming-Xin Lu, Yi-Lei Meng, Sang-Yu Hu, ..., Zhao-Xuan Tian, Fei Yang, Jun-Zhou Wang, Li Li — Briefings in Bioinformatics, 2026-11-05
  SpaMEDM fuses dual-attention graph integration with masked reconstruction and diffusion-based latent regularization to learn robust spatial multi-omics embeddings, improving spatial domain identification under sparsity and noise.

• 🔭 EnzyREx: deep learning enables reaction-based comprehensive enzyme screening
  Hirota, Keisuke, Yamada, Takuji — Genome Biology, 2026-11-03
  EnzyREx combines deep learning enzyme/non-enzyme classification with reaction-capability prediction to screen orphan reactions, identifying candidates for 897 microbiome-related reactions and yielding novel cytotoxic compounds via model-guided prioritization.

• 📰 LEEN: local equivariant energy networks with evolution-guided structural reasoning for protein stability prediction
  Shivram, A, Chivukula, Aneesh Sreevallabh, Gupta, Manik, Chowdhury, Sourav — BMC Bioinformatics, 2026-11-03
  LEEN is a compact, energy-based E(n)-equivariant model that predicts protein stability changes (ΔΔG) from local geometric neighborhoods with evolutionary gating, ensuring anti-symmetry and competitive accuracy using only ~2–3M parameters.

• 📰 Predicting protein-RNA binding signals with a hybrid deep neural network
  Zhou, Yuebo, Ding, Xiaohan, ..., Pan, Xiaoyong, Fan, Yongxian — BMC Bioinformatics, 2026-11-03
  RBPSignal predicts RNA-binding protein binding signals from sequence using deep learning with Integrated Gradients for motif syntax, outperforming baselines and recovering biologically grounded RBP motifs.

• 📰 Deciphering the regulatory code of RNA inosine through enzymatic precision mapping and an explainable deep learning model
  Han, Shaoqing, Li, Jiehua, ..., Zhou, Xiang, Weng, Xiaocheng — Cell Discovery, 2026-11-04
  Combines REX-seq (EndoV+XRN-1) with InoDREAM, an interpretable deep model, to map >590k A-to-I RNA editing sites and predict regulatory proteins, cis-elements, and disease-linked mutations governing selective inosine placement.

• 📰 Structured multi-view relational learning for drug response prediction.
  Zeng-Wei Wang, Wan-Chen Li, Jun-Lin Xu, Ya-Jie Meng, Hao Zhou, Shu-Ting Jin — Briefings in bioinformatics, 2026-11-07
  MulDR preserves omics-view-specific cell-line relations and conditions fusion on multi-representation drug features to predict drug response, improving RMSE and generalizing in cold-start settings.

• 📰 HMVPS-DRP: Hybrid multi-view representation learning and pathway-state-aware graph learning for interpretable cancer drug response prediction with multi-omics data.
  Wei Wang, Di Wang, Biao Li, Yun Zhou — Computational biology and chemistry, 2026-11-04
  HMVPS-DRP integrates hybrid multi-view drug representations with pathway-state-aware heterogeneous graphs and condition-aware aggregation to predict cancer drug response, offering pathway/gene-level interpretability and strong performance on GDSC/PRISM.

• 📰 Evolution-Inspired Multi-Objective Bayesian Optimization for Protein Engineering.
  Kai Wen, Sirui Wang, Yixin Sun, Shiwen Li, Mengsong Wang, Haoyang Liu, Quanshun Li, Jingxuan Zhu — Angewandte Chemie (International ed. in English), 2026-11-08
  EvoMOBO is an evolution-inspired multi-objective Bayesian optimization framework that exploits path-dependent sequence generation and mechanism-derived descriptors to efficiently engineer enzymes with improved conversion and stability.
  Affiliations: School of Life Sciences, Key Laboratory for Molecular Enzymology and Engineering of Ministry of Education

• 📰 Machine-learning-guided discovery of selective CK2 inhibitors with reduced CDK off-target activity
  Ying, Huiyan, Kong, Weikaixin, ..., Aittokallio, Tero, Miihkinen, Mitro — Communications Chemistry, 2026-11-07
  Demonstrates an ML-guided pipeline (with error prediction) to redesign CK2 inhibitors with reduced CDK off-target activity, achieving a 90% hit rate and identifying selective submicromolar allosteric antagonists.

• 📰 A global survey of type I secretion systems and the substrates in Gram-negative bacteria.
  Jia-Lin Li, Ze-Fan Chen, Ke-Xin Liu, Jiang Zhang, ..., Xin-Yue Hu, Xin Cao, Xiao-Min Liu, Ye-Jun Wang — Microbiological research, 2026-11-05
  Performs a global survey of type I secretion systems and substrates, training a Mamba-based model (T1SEstacker 3.0) that outperforms existing tools for T1SS substrate prediction and maps their bacterial distribution.

• 📰 AlphaBridge: Tools for the analysis of predicted biomolecular complexes.
  Daniel Álvarez-Salmoral, Razvan Borza, Carmen Maiella, Bjørn P Y Kwee, Ren Xie, Robbie P Joosten, Maarten L Hekkelman, Anastassis Perrakis — Structure (London, England : 1993), 2026-11-05
  AlphaBridge is an automated toolkit that clusters AlphaFold3-derived interaction motifs, scores binary/multicomponent complexes, and visualizes confidence-bounded interfaces to validate AI-predicted biomolecular assemblies.
  Affiliations: Oncode Institute and the Netherlands Cancer Institute Division of Biochemistry, Amsterdam 1066CX; Oncode Institute and the Netherlands Cancer Institute Division of Biochemistry, Amsterdam 1066CX

• 💻 Multi-view state-space modeling for zero-shot circRNA–miRNA interaction prediction
  Mengmeng Wei, Lei Wang, Yu-An Huang, Zhian Huang, Peng-Wei Hu, Zhu-Hong You — Bioinformatics, 2026-11-05
  MV-S2M models circRNA–miRNA interactions via multi-view state-space learning with structure-aware diffusion and disease-mediated semantic subgraphs, enabling zero-shot CMI prediction and case-supported biological insights.

• 📰 GraphPepTox: a structure-aware pre-trained framework for peptide toxicity prediction using geometric graph transformer and LoRA-based fine-tuning
  Bhaduri, Soumyadeep, Das, Debraj, Mitra, Pralay — BMC Bioinformatics, 2026-11-06
  GraphPepTox introduces a pre-trained geometric graph transformer with masked residue reconstruction for peptide structures, improving toxicity prediction by explicitly modeling 3D residue interactions and conformational organization.

🧬 Preprints (arXiv + bioRxiv)

82 matched filters -> 20 selected after LLM relevance + novelty ranking.

• 📄 ProteoCraft: Protein Design as Proposal-Verification Search
  Fang Wu, Hongbin Lin, Bingchen Duan, Yuchen Wang, ..., Shuangjia Zheng, Yao Shu, Jure Leskovec, Yejin Choi — arXiv, 2026-11-02
  Introduces a proposal–verification tree search that uses approximate structural gradients from diffusion-based structure predictors to guide protein sequence edits, then verifies with no-gradient folding, yielding higher-quality de novo binders and 82.9% experimental hit rate (PD-L1, IFNAR2).

• 🧬 Accelerating antibiotic discovery from Earth's isolated habitats through protein language models and microfluidics
  Kang, Z.; Zhou, Q.; Zhang, H.; Liu, J.; ...; Wang, Y.; Chen, P.; Liu, B.-F.; Ning, K. — bioRxiv, 2026-11-04
  Combines a protein language model ensemble (SPEAR) with a microfluidic screening platform (MicroGrad) to discover antimicrobial peptides from metagenomes, validating 79/85 candidates in vitro and four broad-spectrum leads in a mouse infection model.
  Affiliations: Huazhong University of Science and Technology

• 🧬 Few-shot regulatory sequence design with a genomic language model
  Hao, Y. A.; Liu, M.; Nori, D.; Brixi, G.; ...; Chang, D.; Park, R.; Allen, W. E.; Hie, B. — bioRxiv, 2026-11-04
  Demonstrates generative in-context learning in a genomic language model (Evo 2) to design regulatory DNA from few examples, with in vitro enhancer activity across cell lines and in vivo neuron-subtype–specific labeling in mouse cortex.
  Affiliations: Stanford University

• 📄 Empirical Bayes Flow Matching for Continuous Cryo-EM Heterogeneity
  Daniel Aibinder, Azmi Haider, Dan Rosenbaum — arXiv, 2026-11-03
  Proposes Empirical Bayes Flow Matching to learn generative priors over protein conformations directly from cryo-EM projections via a stochastic-approximation EM loop, recovering continuous heterogeneity in atomic coordinate space without clean pretraining.

• 📄 Predicting protein structures at any folding stage from sequence
  Lisheng Zhang, Qingsong Li, Zinan Xie, Yanting Yang, ..., Zhican Fu, Wentao Cui, Runtong Hou, Zhengwei Xie — arXiv, 2026-11-03
  FoldScape conditions on folding progress to generate protein conformational ensembles from unfolded to native states, covering experimental folding observables far better than seed-varied AlphaFold 3 ensembles with controllable Q conditioning.

• 📄 iGC: Antibody Optimization via Reinforcement Learning over a Learned Somatic Hypermutation Prior
  Thao Nguyen, Robert G. Alberstein, Ewa Nowara, Homa MohammadiPeyhani, ..., Franziska Seeger, Andrew Leaver-Fay, Richard Bonneau, Joshua Yao-Yu Lin — arXiv, 2026-11-03
  iGC combines a learned somatic hypermutation prior with KL-regularized reinforcement learning and a reward-tax meta-objective to optimize antigen-specific antibody affinity while preserving realistic mutation patterns and diversity.

• 📄 Probabilistic Steering of Generative Priors for Lab-in-the-Loop Antibody Optimization
  Anisha Parsan, Sebastian Ober, Calvin McCarter, Nick Bhattacharya, Hunter Elliott — arXiv, 2026-11-02
  Unifies discovery and optimization by probabilistically steering a structure-based generative prior with a surrogate’s posterior uncertainty, improving low-N lab-in-the-loop antibody optimization and maintaining design diversity.

• 🧬 A protein language model for domain, functional site, and variant effect prediction
  Wang, P.; Ahsan, M. U.; Chen, T.; Wang, K. — bioRxiv, 2026-11-05
  PALM integrates ESM2 with LoRA, layer aggregation, and BiLSTM to jointly predict domains and functional sites and to score missense effects, outperforming ESM baselines and rivaling AlphaMissense in zero-shot variant classification.
  Affiliations: Children's Hospital of Philadelphia

• 🧬 Zero-Shot Protein-Ligand Binding Site Prediction from Protein Sequence and SMILES
  Pourmirzaei, M.; Alqarghuli, S.; Chen, K.; Pourmirzaei, M.; Xu, D. — bioRxiv, 2026-11-07
  Presents a ligand-aware benchmark and a sequence+SMILES model that conditions a PLM on CLM embeddings to predict residue-level binding sites, enabling zero-shot generalization to unseen ligands with competitive performance.
  Affiliations: University of Missouri

• 🧬 Omics-scale decoding of protein N-glycosylation using protein language model embeddings
  Zhou, X.; Guan, X.; Zhou, Y.; Wang, J.; ...; Ren, W.; Wang, W.; Xie, R.; Li, W. — bioRxiv, 2026-11-04
  MaskNGly decodes N-glycosylation propensities directly from PLM embeddings via coevolutionary analysis, achieving zero-shot site prediction at proteome scale and revealing evolutionary and structural trends in glycosylation.
  Affiliations: Nanjing University

• 📄 RangeFinder: Predicting Ligand Property Ranges from Protein Pockets
  Alan Nadelsticher Ruvalcaba, Chris Alvin, Supratik Mukhopadhyay, Adam Bess — arXiv, 2026-11-02
  RangeFinder uses an SE(3)-equivariant transformer to predict pocket-conditioned ranges over ligand physicochemical properties, substantially shrinking chemical search space while retaining compatible binders.

• 🧬 Large-scale AI-guided discovery of antifreeze peptides reveals functional tandem-repeat architectures
  Li, Y.; Wang, C.; Liu, J.; Song, M.; ...; Yao, J.; Zhao, L.; Yu, H.; Wang, X. — bioRxiv, 2026-11-04
  An AI-guided pipeline fusing multimodal AFP prediction, sequence generation, and repeat-pattern mining identifies antifreeze peptide families, with multiple short peptides validated for ice recrystallization inhibition and cryoprotection in cells.
  Affiliations: Beijing Key Laboratory of Artiffcial Intelligence for Advanced Chemical Engineering Materials, State Key Laboratory of Chemical Engineering and Low-Carbon Techn

• 📄 Multi-Objective Generative Optimization via Expected Hypervolume Fine-Tuning
  Juan Guevara, Riccardo De Santi, Andreas Krause, Raul Astudillo — arXiv, 2026-11-03
  Hypervolume Fine-Tuning maximizes expected Pareto hypervolume of samples from pretrained generators, using unbiased marginal hypervolume rewards to achieve state-of-the-art multi-objective coverage in peptide and small-molecule design.

• 📄 CoMoFlow: Controllable Protein Conformational Ensemble Generation via Explicit Collective-Mode Modeling
  Yusong Wang, Zheyuan Gu, Ailiang Lin, Minghao Shao, Yicheng Xu, Jiang Cai, Prayag Tiwari, Mingkun Xu — arXiv, 2026-11-06
  CoMoFlow explicitly models protein-specific collective modes to condition SE(3) flow matching, improving reconstruction of MD conformational distributions and enabling targeted control of backbone motions.

• 📄 Static Recovery Is Not Dynamic Stability: Dynamics-Aware Benchmarking of Protein Motif Scaffolding
  Emile de Bruyn, Moritz Schäffler, Jannik Schneider, Sebastian R. Schmidt, ..., Alexander Schug, Stefan Kesselheim, Artur Yakimovich, Alina Bazarova — arXiv, 2026-11-05
  Introduces a large, systematic motif-scaffolding benchmark (BILBY-PRO) and an MD-based evaluation (BILBY-MD) showing that static refolding success often lacks dynamical fidelity, providing more rigorous criteria for functional scaffold generation.

• 📄 El Agente Fármaco: A Scalable AI Scientist for Drug Discovery
  Marcel Müller, Mohammad Ghazi Vakili, Yoshihiro Ogura, Ella Ashoori, ..., Yunheng Zou, Abraham C. Stern, Varinia Bernales, Alán Aspuru-Guzik — arXiv, 2026-11-02
  El Agente Fármaco is an LLM-driven agent that composes end-to-end virtual screening workflows with a provenance-tracked scientific memory, proposing drug-like candidates efficiently and competitively under explicit multi-criteria constraints.

• 📄 HistoTx: Early fusion of H&E images and spatial transcriptomics at varying spatial transcriptomics resolution with self-supervised learning
  Alice Driessen, Emil Malmsten, Steffen Wolf, Bernd Illing, ..., Krisztian Homicsko, Raphael Gottardo, Marianna Rapsomaniki, Fei Tang — arXiv, 2026-11-02
  HistoTx early-fuses image and spatial-transcript tokens in a shared transformer, preserving pathology-foundation performance on image-only inference while improving gene-expression prediction and multimodal tasks when both inputs are available.

• 📄 The DynaCell Evaluation Suite for Dynamic 3D Virtual Staining of Live Cells
  Alexandr A. Kalinin, Dihan Zheng, Taylla Milena Theodoro, Ivan Ivanov, ..., Manuel D. Leonetti, Carolina Arias, Bo Huang, Shalin B. Mehta — arXiv, 2026-11-02
  Provides a 3D time-lapse virtual-staining evaluation suite with curated datasets, baselines, and phenotype-tiered metrics, revealing trade-offs between pixel fidelity and phenotypic profiling for live-cell imaging models.

• 📄 BEND 2.0: Benchmarking the Capabilities Needed to Design DNA
  Andreas Bjerregaard, Rachael DeVries, Dionysia Danai Brilli, Panagiotis Antoniadis, ..., Anders Krogh, Wouter Boomsma, Ole Winther, Frederikke Isa Marin — arXiv, 2026-11-02
  BEND 2.0 benchmarks DNA language models on prediction and design across eight tasks, showing that linear readouts on frozen representations can rival native heads and exposing gaps in context-specific and multi-cell design capabilities.

• 📄 Molecular Prior Networks
  Pawel Gomoluch, Anthony Bradley — arXiv, 2026-11-03
  Molecular Prior Networks meta-learn a transformer to predict query affinities from labeled supports in a single forward pass across thousands of datasets, enabling fast few-shot drug–target affinity modeling.