Weekly BioML Digest [November 02, 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)

1712 matched filters -> 20 selected after LLM relevance + novelty ranking.

• 📰 DreamFold: A World Model to efficiently generate protein folding pathways in the latent space.
  Alan Ianeselli, Jewon Im, Eddie Cavallin, Mark B Gerstein — Protein science : a publication of the Protein Society, 2026-11-01
  A latent world model of protein folding dynamics enables learned planning of dihedral moves and O(N) cost scaling, exposing therapeutic cryptic sites missed by static methods.
  Affiliations: Department of Molecular Biophysics and Biochemistry, Yale University; Department of Computer Science, Yale University; ...

• 📡 IDPForge: deep learning of proteins with global and local regions of disorder
  Castro, Stefano, Zhang, Oufan, ..., Forman-Kay, Julie D., Head-Gordon, Teresa — Nature Communications, 2026-10-30
  Presents IDPForge, a transformer language model–driven diffusion generator that produces all-atom ensembles for intrinsically disordered proteins/regions consistent with NMR/SAXS, bridging a key gap beyond static predictors.

• 📰 Enhanced diffusion sampling: Efficient rare event sampling and free energy calculation with diffusion models.
  Yu Xie, Ludwig Winkler, Lixin Sun, Sarah Lewis, ..., Iryna Zaporozhets, Cecilia Clementi, Christopher M Bishop, Frank Noé — The Journal of chemical physics, 2026-10-28
  Develops enhanced diffusion sampling (UmbrellaDiff/MetaDiff/ΔG-Diff) to steer diffusion-model equilibrium samplers for rare-event exploration and unbiased free-energy estimation, enabling protein folding ΔG in minutes–hours.
  Affiliations: Microsoft Research AI for Science, Berlin; Microsoft Research AI for Science, Cambridge; ...

• 📰 Ernie-Switcher: RNA foundation model-guided design of programmable RNA devices in a navigable landscape.
  Xin Lei, Runting Li, Yuzhong Liu, Jiesheng Huang, ..., Shimin Liu, Pengxia Wang, Huaiguang Jiang, Wei Xiong — Bioinformatics (Oxford, England), 2026-10-26
  RNAErnie-guided switch optimization leverages a high-navigability landscape and constraints to produce high-ON/OFF toeholds validated in vivo.
  Affiliations: School of Future Technology, South China University of Technology; Guangdong Engineering Research Center of Low-Carbon Synthetic Biotechnology, Guangzhou 511442; ...

• 📰 Integrated AI-Guided de novo CDR-H3 Design and Combinatorial Phage Display for Antibody Discovery
  Kairi Furui, Shinji Soga, R. Yamazaki, S. Yagi, ..., Takatsugu Kosugi, T. Shimomura, Kana Gondo, Masahito Ohue — Antibody Therapeutics, 2026-10-27
  Demonstrates SANDAN, an AI-guided, structure-aware de novo CDR-H3 design workflow combined with phage display to discover hTIGIT-binding antibodies competing with a reference epitope without known complex poses.

• 📰 An extensible deep learning framework for multi-type chromatin interaction prediction through hierarchical adaptation strategies.
  Shuai-Bin Wang, Tong Chen, Wei-Xu Liu, Xue-Jie Zheng, Zhong-Xin Yang, Zhen Liang, Yin Shen — PLoS computational biology, 2026-10-27
  UniLoopNet (multi-task DL) consistently predicts enhancer–promoter, enhancer–enhancer, CTCF loops, and stripes from sequence, retaining performance under realistic imbalanced and locus-disjoint evaluations.

• 📰 TFXcan reveals transcription factor programs driving GWAS loci.
  Temidayo Adeluwa, Sofia Salazar-Magaña, Sarah Sumner, Saideep Gona, ..., Matthew Stephens, Guimin Gao, Tiffany Amariuta, Hae Kyung Im — Cell genomics, 2026-10-27
  TFXcan associates genetically predicted TF binding with traits via Enpact and matrix decomposition, revealing TF–tissue programs that drive risk at GWAS loci.
  Affiliations: Committee on Genetics, Genomics; Department of Medicine, Section of Genetic Medicine; ...

• 📰 ΔNAzyme: A Foundational QM/MM Δ-Machine Learning Potential for Nucleic Acid Enzyme Catalysis.
  Erika McCarthy, Timothy J Giese, Şölen Ekesan, Julie Puyo-Fourtine, Lauren Lerew, Andreas W Götz, Darrin M York — Journal of chemical theory and computation, 2026-10-27
  A Δ-MLP corrects DFTB3 QM/MM to hybrid DFT for NAzyme reactions, transferring across classes and enabling accurate, efficient free-energy profiling with flexible fine-tuning.
  Affiliations: Laboratory for Biomolecular Simulation Research, Institute for Quantitative Biomedicine and Department of Chemistry and Chemical Biology; San Diego Supercomputer Center, University of California San Diego

• 🖥️ Machine learning dataset and benchmark for accurate T cell receptor–pHLA binding prediction
  Zhu, Xinyuan, Lu, Jiadong, ..., Zhang, Yuyan, Feng, Fuli — Nature Computational Science, 2026-10-26
  Releases Hi-TPH, a hierarchical TCR–pHLA binding dataset and fine-tuned PLM suite with improved negative sampling, delivering 19% gains over SOTA on an external wet-lab test and dissecting component contributions.

• 📡 Multiproperty optimization of therapeutic antibodies through sequence editing via large language model and reinforcement learning
  Qiu, Minyao, Wang, Zhiwei, ..., Xiong, Zhankun, Zhang, Wen — Nature Communications, 2026-10-26
  AbOptimizer combines RL with protein language models and Pareto rewards to edit antibody sequences for multi-property optimization, yielding in vitro gains in affinity, specificity, stability, and indel-inclusive designs.

• 📰 SprintX: mitigating data scarcity in antibody design via synthetic structural prior knowledge dataset.
  Ryong Heo, Dahyeon Lee, Jaeho Jung, Byung Ju Kim, Chihyun Park — mAbs, 2026-10-30
  SprintX is an AlphaFold3-derived synthetic dataset of antibody–antigen complexes with rigorous QC that enhances pretraining for antibody generation, binding affinity, and interface prediction.

• 📰 SAbDab2: the structural antibody database in the age of machine learning.
  Henriette L Capel, Odysseas Vavourakis, Benjamin H Williams, Christopher R Taylor, Charlotte M Deane — Nucleic acids research, 2026-10-31
  SAbDab2 modernizes the structural antibody database with ML-grade standardized structures and versioned train/test splits across formats, enabling reproducible ML development for antibody modeling.
  Affiliations: Department of Statistics, University of Oxford

• 📰 GCMDiff: a condition-guided diffusion framework for pharmacophore-fragment drug discovery and target-specific molecular design
  Wei, Ya-Ko, Lee, Jung-Yu, ..., Chu, Jhih-Wei, Yang, Jinn-Moon — Journal of Cheminformatics, 2026-10-30
  GCMDiff’s interpretable condition embedding steers graph diffusion to generate molecules satisfying atom counts, Lipinski rules, and pharmacophore fragments, with practical success on EGFR inhibitors.

• 📰 Interpretable Drug–Target Prediction via Geometric Alignment: A Linear Alternative to Deep Architectural Complexity
  Aashutosh Ghimire, C. Alvin, Adam Bess, Supratik Mukhopadhyay — Molecular Informatics, 2026-10-30
  Aligns pre-trained ligand/protein embeddings into a shared manifold for SOTA DTI accuracy and cosine-similarity interpretability that mirrors kinase phylogeny.

• 📰 DRAGON-3D: an electron-density-aware generative model for dual-target drug molecule generation.
  Zhong-Liang Zhou, Hong-Xing Zhang, Shi Feng, Yan Zhao, Zhen-Yang Liu, Jun-Bao Huang, Qing-Chuan Zheng — Physical chemistry chemical physics : PCCP, 2026-10-27
  Electron-density–aware latent models generate dual-target ligands from single-target pretraining, improving generalizability and interpretability of dual-pocket compatibility.

• 📰 Latent Style-Based Quantum Wasserstein GAN for Drug Design.
  Julien Baglio, Yacine Haddad, Richard Polifka — Journal of chemical information and modeling, 2026-10-26
  A hybrid quantum GAN with a small generator competes with classical baselines on MOSES and demonstrates scalable inference on quantum hardware.
  Affiliations: Center for Quantum Computing and Quantum Coherence (QC2), Department of Physics; QuantumBasel , Schorenweg 44b; ...

• 📰 AI-driven photophysics-aware design of fluorescent probes with applications in α-synuclein biosensing and inhibitor screening.
  Xuefeng Jiang, Ruina Luo, Cenxing Zhou, Hailun Deng, ..., Yi Zou, Zhenhua Liu, Xuan Zhai, Biyue Zhu — Biosensors & bioelectronics, 2026-11-01
  A photophysics-aware generative framework co-optimizes binding and fluorescent properties, yielding an α-synuclein turn-on probe and enabling inhibitor HTS hit identification.
  Affiliations: Department of Pharmacy, Department of Neurosurgery; Basic Medicine Research and Innovation Center for Novel Target and Therapeutic Intervention (Ministry of Education), College of Pharmacy; ...

• 📰 A deep learning and generative modeling pipeline for mining and engineering alkaline-stsable xylanases.
  Ruohan Zhang, Yiyang Zhang, Zhonghao Deng, Pengtao Wang, ..., Jian Tian, Huoqing Huang, Ming Li, Feifei Guan — Bioresource technology, 2026-11-01
  Deep generative diversification and screening yield an alkali-stable enzyme variant with shifted pH optimum and improved activity for industrial applications.
  Affiliations: State Key Laboratory of Animal Nutrition and Feeding, Institute of Animal Science; National Key Laboratory of Agricultural Microbiology, Biotechnology Research Institute; ...

• 📰 Discovering hidden candidate plastic-degrading enzymes: Combined multi-omics and machine learning strategy.
  Flavio Agostini, Virginia Baruzzo, Feliciana Real Fernandez, Alessandro Satta, ..., Maria Chiara Valerin, Stefano Campanaro, Laura Treu, Guido Zampieri — Bioresource technology, 2026-11-01
  Semi-supervised ranking across multi-omics enriches for plastic-degrading enzymes beyond sequence hits, supported by expression and structural similarity.
  Affiliations: Department of Biology, University of Padova; Institute of Chemistry of Organometallic Compounds, National Research Council (ICCOM-CNR); ...

• 📰 EVLncRNAs 4.0: an updated database of experimentally validated functional long non-coding RNAs with evidence-grounded AI access.
  Bai-Ling Zhou, Bao-Hua Ji, Zhi-Wei Li, Zheng-Hua Li, ..., Hui-Ying Zhao, Yue-Dong Yang, Yao-Qi Zhou, Ji-Hua Wang — Nucleic acids research, 2026-10-31
  EVLncRNAs 4.0 scales curated functional lncRNA evidence and deploys RAG-based AI that composes answers strictly from retrieved, PMID-linked records for robust, context-grounded querying.

🧬 Preprints (arXiv + bioRxiv)

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

• 🧬 Design of full-length agonist and antagonist antibodies for hard targets
  Pythia Labs Team,; Aglietti, R. A.; Ansari, R.; Bowers, P. M.; ...; Sosna, J.; Tacke, R.; Yeates, T. O.; Zanini, L. — bioRxiv, 2026-10-26
  Pythia-1, an all-atom generative model conditioned on explicit design constraints, designs full-length IgG antibodies with specified epitopes and states, yielding nanomolar GPCR agonists/antagonists and broadly developable leads. This end-to-end pipeline demonstrates programmable antibody function across 22 hard targets from a single design round.
  Affiliations: Pythia Labs

• 📄 Accurate structural modelling of protein-glycan recognition with GlycoWeave
  Huifeng Zhao, Jintu Zhang, Yifei Xiao, Youqiao Qian, ..., Zhongjun Li, Zhongtang Li, Tingjun Hou, Yu Kang — arXiv, 2026-10-27
  GlycoWeave introduces a glycan-specific diffusion framework with internal-coordinate sampling and equivariant multiscale representations to model protein–glycan complexes. It vastly outperforms AlphaFold 3 while preserving glycan stereochemistry and linkage conformations across diverse architectures.

• 🧬 Predicting conformational changes induced by biomolecular interactions
  Bae, Y.; Baek, J.; Choi, E.; Choi, J. H.; ...; Kim, H. M.; Kim, W. Y.; Lee, G. R.; Oh, B.-H. — bioRxiv, 2026-10-28
  K-Fold formulates complex assembly as a diffusion bridge from unbound monomer priors to bound states, capturing interaction-induced conformational changes without MSAs. It accurately recovers non-coevolved interfaces in antibodies/degraders and regulatory states in GPCRs and kinases.
  Affiliations: Department of Chemistry; Kim Jaechul Graduate School of AI; ...

• 🧬 A high-resolution protein-RNA binding atlas for structure-aware interaction prediction and post-transcriptional regulatory modeling
  Karimzadeh, M.; Fish, L.; Karner, H.; Linder, J.; ...; Wang, B.; Najafabadi, H. S.; Leonetti, M.; Goodarzi, H. — bioRxiv, 2026-10-29
  A high-throughput ecCLIP atlas of 67 RBPs and Pythia, a structure-aware neural model with grammar-inspired convolutions, jointly enable high-resolution RBP–RNA interaction prediction and variant-effect interpretation. MPRC assays validate Pythia’s decomposition of sequence- versus structure-dependent binding.
  Affiliations: Arc Institute, University of California

• 🧬 Few-shot regulatory sequence design with a genomic language model
  Hie, B.; Hao, Y. A.; Liu, M.; Nori, D.; ...; Ramakrishnan, C. A.; Chang, D.; Park, R.; Allen, W. E. — bioRxiv, 2026-10-30
  A genomic LLM (Evo 2) exhibits few-shot in-context learning to generate cell-type-specific regulatory DNA, producing enhancers validated in vitro and in vivo. Reinforcement learning with sequence-to-function rewards further strengthens conditioning for generative biological design.
  Affiliations: Stanford University

• 🧬 Efficient biological design via genomic in-context learning
  Brixi, G.; Hao, Y. A.; Nori, D.; Merchant, A.; ...; King, S.; Shalvarjian, K.; Park, D.; Hie, B. — bioRxiv, 2026-10-29
  Genomic in-context learning with Evo 2 enables de novo generation of metal-binding proteins, antibodies, protein–RNA systems, and complete bacteriophage genomes, improved via a ‘genomic chain-of-thought’ prompting strategy. An efficient inference engine scales context to megabases for rapid iteration.
  Affiliations: Stanford University

• 🧬 MultiFlow: coupled flow matching for predicting single-cell multiomic perturbation responses in unseen cellular contexts
  Wang, H.; Zhang, C.; Zhang, M.; Nie, X.; Liu, Q. — bioRxiv, 2026-10-28
  MultiFlow couples flow-matched generative models for RNA and ATAC with bidirectional information exchange to learn joint multiomic states and predict perturbation responses. It generalizes to unseen cell types while preserving cross-modal coordination such as peak–gene effects.
  Affiliations: Yale University

• 🧬 Reconstructing human gene regulatory evolution using deep learning
  Yamada, T.; Yuan, X.; Zeberg, H.; Soltys, V.; Lana Alberro, M.; Sarropoulos, I.; Paabo, S.; Kaessmann, H. — bioRxiv, 2026-10-31
  AlphaGenome, a sequence-to-function deep model, predicts cis-regulatory effects of ~17M human-lineage substitutions across 305 tissues, mapping regulatory gains and linking them to expression shifts. It pinpoints causal motif-creating variants and traces their timing relative to archaic hominins.
  Affiliations: Center for Molecular Biology of Heidelberg University (ZMBH)

• 📄 TACFormer: Transformer-Based Generative Model for Target-Conditioned De Novo PROTAC Design
  Yossra Gharbi, Rocío Mercado — arXiv, 2026-10-29
  TACFormer conditions a GPT-style molecular generator on pretrained ESM-2 protein embeddings to design de novo PROTACs. It steers full-molecule generation (not just linkers) toward target-aware degrader chemotypes with high novelty and diversity.

• 📄 D2ML: Models for Direct-to-Biology Small Molecule Screening
  AkshatKumar Nigam, Neri Amara, Hillary Dequina, Dyana Kenanova, ..., Raymond V. Fucini, Jack Sadowsky, Stig K. Hansen, Imran S. Haque — arXiv, 2026-10-29
  Assay-aware neural models leverage noisy direct-to-biology screening data (single-point, crude dose-response, purified) to jointly predict activity and selectivity. Prospective selection improved the potency–selectivity Pareto front for ROS1 over TRKB within rapid D2B cycles.

• 📄 Locality-Controlled Out-of-Distribution Guidance for Robust Regulatory DNA Sequence Design
  Dabin Jeong, Marwane Bourdim, Mo Lotfollahi — arXiv, 2026-10-30
  LOCOGEN introduces locality-controlled guidance for regulatory DNA diffusion by learning motif-invariant masks and damping only background positions, avoiding reward hacking. It jointly optimizes property, novelty, and plausibility across multiple human cell lines.

• 📄 Towards Generalizing Pretrained All-Atom Representations to Unified Binder Design
  Maojiang Su, Jiale Chen, Qijie Zhu, Shang Wu, ..., Jihai Zhao, Guo Ye, Yibo Wen, Han Liu — arXiv, 2026-10-27
  PairFuse fuses pretrained all-atom representations into a geometric latent diffusion model via parameter-efficient adapters, enriching block features and pairwise structure in the denoiser. It boosts unified binder design performance across antibodies, peptides, and small molecules.

• 🧬 Illuminating phage dark matter beyond the reach of structural alignments with PhageFlare
  Wang, C.; Xu, Y.; Li, Z.; Zhang, L.; Zhang, L. — bioRxiv, 2026-10-29
  PhageFlare combines protein language-model embeddings with deep classifiers to annotate phage proteins lacking structural homologs, outperforming structural-alignment pipelines under structure-disjoint constraints. It expands functional coverage of metagenomic viral dark matter and flags lysogenic potential.
  Affiliations: Hong Kong Baptist University

• 🧬 Can Folding Models Tell Binders from Bluffers? Evidence from POISK: The Patent-Derived Antibody Dataset
  Tupikina, D.; Roncoli, A.; Bujotzek, A.; Abanades, B. — bioRxiv, 2026-10-26
  POISK, a large patent-derived antibody–antigen dataset, benchmarks folding models on binder discrimination vs decoys lacking structural data. State-of-the-art predictors enrich for true binders but remain limited in epitope localization and single-candidate identification.
  Affiliations: Prescient Design - AI for Drug Discovery, Computational Science Center Of Excellence

• 📄 Transferring Peptide-Target Interaction Patterns to Small Molecules by Joint Guidance for Diffusion and Flow Matching
  Jian Lee, Changyun Cho, Sun Kim — arXiv, 2026-10-30
  PIFT transfers peptide–target interaction patterns to small-molecule generation by jointly guiding masked discrete diffusion and continuous flow matching. It improves interface recovery and docking on PPI targets while preserving a strong molecular prior.

• 🧬 Multi-resolution humanness scoring of antibodies via a region-isolated language model
  Liu, H.; Chen, P.; Zhou, S.; Wei, B.; ...; Liu, X.; Zhou, S.; Han, L.; Ye, L. — bioRxiv, 2026-10-28
  IgScope is a region-isolated antibody language model that scores humanness at chain, IMGT-region, and residue resolution, detecting murine back-mutations and correlating with clinical ADA incidence. It audits de novo designs by exposing non-human-like CDRs concealed by whole-chain scores.
  Affiliations: School of Pharmacy, Faculty of Medicine & Laboratory of Drug Discovery from Natural Resources and Industrialization

• 📄 From WSIs to Bulk, Spatial and Single-Cell Transcriptomics via Multi-Scale Graph Neural Networks
  Miguel Andrade, José Almeida, Carlos Santiago, Catarina Barata — arXiv, 2026-10-30
  A multi-scale GNN predicts bulk, spatial, and single-cell transcriptomes directly from WSIs using PFM features and patch/cell graphs with cross-scale consistency. It enables survival analysis, spatial hazard mapping, and tumor classification from routine histology.

• 🧬 BLOOM: Biologically Informed Flow Matching for Transcriptomic Virtual Embryo Modeling
  Liang, Z.; Li, S.; Li, B.; Huang, Y.; Yu, L. — bioRxiv, 2026-10-27
  BLOOM uses biologically informed conditional flow matching to generate developmental single-cell populations at queried times, accurately interpolating and rolling out trajectories. It recovers program timing and regulatory associations and probes gene perturbations across time.
  Affiliations: The University of Hong Kong

• 🧬 ProtCell: a foundation model for single-cell proteomics
  Hu, B.; Li, Z.; Jiang, Q.; Yin, Q.; Liu, Z.; Jiang, R. — bioRxiv, 2026-10-27
  ProtCell is a foundation transformer trained on ~380M single-cell proteomics profiles that transfers annotations, reconstructs unmeasured proteins, and infers latent cytotoxic programs. Learned protein dependencies guide compact panel design while preserving biological organization.
  Affiliations: Tsinghua University

• 📄 EVA-Flow: Learning Shared Chemistry for Unified Environment-Aware Molecular Conformation Generation
  Bing Yan, Anuroop Sriram, Zachary Ward Ulissi, Kyunghyun Cho, Ricky T. Q. Chen, Benjamin Kurt Miller — arXiv, 2026-10-26
  EVA-Flow learns shared chemistry via pretraining and conditions on environment to generate conformations across vacuum, docking, solvation, and crystals. It provides the largest gains in data-scarce settings and captures environment-specific conformer diversity.