Weekly BioML Digest [August 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)

1288 matched filters -> 20 selected after LLM relevance + novelty ranking.

• ⚗️ Generative design of bacteriophages with genome language models.
  Samuel H. King, Claudia L. Driscoll, David B. Li, Daniel Guo, Aditi T. Merchant, G. Brixi, Max E. Wilkinson, Brian L. Hie — Science, 2026-08-06
  Trains genome-scale language models to generate viable bacteriophages with specified host tropism, demonstrating experimental infectivity and therapy potential.

• 🌿 Zero-shot design of drug-binding proteins via neural iterative selection−expansion
  Fry, Benjamin, Slaw, Kaia, Polizzi, Nicholas F. — Nature, 2026-08-06
  Pairs a ligand-aware GNN (LASErMPNN) with a structure predictor in a closed-loop to zero-shot design tight, selective small-molecule-binding proteins with picomolar affinities.

• 🏛️ AI-driven PROTAC design overcomes oncogenic resilience by eliminating the CLIP1-LTK fusion protein.
  Shicheng Chen, Haiting Duan, S. Zhong, Jingxuan Ge, ..., Xiaowu Dong, Jinxin Che, Tingjun Hou, P. Pan — Proceedings of the National Academy of Sciences of the United States of America, 2026-08-06
  Uses deep learning-guided ternary complex modeling and structure-based optimization to design an oral PROTAC that degrades the CLIP1–LTK fusion, overcoming kinase-inhibitor resistance.

• 📡 Rank-guided learning accelerates automated enzyme engineering
  Xu, Jingyi, Zheng, Yan, ..., Yuchi, Zhiguang, Yuan, Yingjin — Nature Communications, 2026-08-03
  Introduces REAP, a closed-loop AI–robotics enzyme engineering platform with a rank-guided learning objective that accelerates discovery of high-activity variants across multiple enzymes.

• 🔬 Expanding the scope of protein language modeling to protein-protein interactions with MSA Pairformer.
  Yo Akiyama, Zhidian Zhang, Olivia Tang, Rachel Seongeun Kim, Milot Mirdita, Martin Steinegger, Sergey Ovchinnikov — Cell, 2026-08-06
  MSA Pairformer extends protein language modeling to complexes, learning interface coevolution and contact patterns from MSAs with strong gains in PPI contact prediction.
  Affiliations: Department of Biology, Massachusetts Institute of Technology; Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology; ...

• 🔬 UniPert-G2CP bridges genetic and chemical screens from molecular representation to phenotype modeling.
  Yiming Li, Min Zeng, Jun Zhu, Linjing Liu, ..., Longkai Huang, Fan Yang, Min Li, Jianhua Yao — Cell, 2026-08-06
  Proposes UniPert-G2CP, a deep framework unifying genetic and chemical perturbation representations to transfer phenotypic effects across modalities and cell types.
  Affiliations: Hunan Provincial Key Laboratory on Bioinformatics, School of Computer Science and Engineering; AI for Life Sciences Lab, Tencent; ...

• 📰 ActivityDiff: A diffusion model with Positive and Negative Activity Guidance for De Novo Drug Design.
  Huimin Zhu, Renyi Zhou, Jing Tang, Min Li — Bioinformatics (Oxford, England), 2026-08-06
  ActivityDiff is a classifier-guided diffusion model that jointly optimizes on-target activity and suppresses off-targets using positive and negative activity guidance for de novo drug design.
  Affiliations: School of Computer Science and Engineering, Central South University; Research Program in Systems Oncology, University of Helsinki

• 📰 READ: A Retrieval-Alignment Diffusion Framework for Structure-based Drug Design.
  Dong Xu, Zhangfan Yang, Junchuang Cai, Sisi Yuan, Zexuan Zhu, Jianqiang Li, Junkai Ji — IEEE transactions on computational biology and bioinformatics, 2026-08-06
  READ conditions SBDD diffusion on retrieved homologous ligands via retrieval–alignment, improving docking-based hit generation with a practical structure-conditioned generative paradigm.

• 📰 High-accuracy structural modeling of antibody-antigen complexes
  Wang, Suhui, Zhuang, Jianan, ..., Hou, Dongliang, Zhang, Guijun — Communications Biology, 2026-08-07
  DeepAAAssembly integrates learned interchain distances with flexibility-aware sampling to assemble antibody–antigen complexes, outperforming AlphaFold3 on DockQ benchmarks.

• 🏛️ Contrastive learning unites sequence and structure in a global representation of protein space.
  Guy Yanai, Gabriel Axel, Liam M Longo, Nir Ben-Tal, Rachel Kolodny — Proceedings of the National Academy of Sciences of the United States of America, 2026-08-11
  CLSS uses contrastive learning to co-embed protein sequences and structures into a shared latent space, improving classification and capturing global sequence–structure relationships.
  Affiliations: Department of Computer Science, University of Haifa; Department of Biochemistry and Molecular Biology, School of Neurobiology; ...

• 📰 PlantCAD2: A DNA foundation model for interpreting genomes across flowering plants.
  Jingjing Zhai, Aaron Gokaslan, Sheng-Kai Hsu, Szu-Ping Chen, ..., M. Romay, Matt Pennell, V. Kuleshov, E. Buckler — Cell genomics, 2026-08-07
  PlantCAD2 is an extended-context DNA foundation model pretrained on 65 plant genomes that outperforms larger models in conservation and regulatory prediction across species.

• 📰 Unified Sampling and Ranking for Protein Docking With DFMDock.
  Lee-Shin Chu, Sudeep Sarma, Da Xu, Jeffrey J Gray — Proteins, 2026-08-06
  DFMDock unifies diffusion-based sampling with an energy head trained via denoising force matching, enabling generation and ranking of protein–protein docked complexes without MSAs.
  Affiliations: Department of Chemical and Biomolecular Engineering, Johns Hopkins University; Program in Molecular Biophysics, Johns Hopkins University; ...

• 📰 Multiscale learning of gene network-driven phenotypic dynamics of single cells
  Zhang, Dongyan, Li, Jinan, Nie, Qing, Sun, Xiaoqiang — Molecular Systems Biology, 2026-08-03
  GRNvelo couples PINNs with gene regulatory networks to infer single-cell velocities, latent time, and multiscale phenotypic dynamics, enabling perturbation prediction.

• 📰 Modeling antibody recognition across SARS-CoV-2 and SARS-CoV-1 using epitope-informed transfer learning
  Jimmy Yuan, Ryan Bruneau, Stephen Won, Aidan Elliot Heller, Thomas Sheffield, Kenneth Sale, Brooke Harmon, L. Pham — Frontiers in Immunology, 2026-08-05
  Uses epitope-centered encoding and transfer learning from SARS‑CoV‑2 to predict antibody affinities to SARS‑CoV‑1 variants with limited target-specific data, validated by ELISA.

• 📰 Multiscale higher-order molecular simplicial complex embedding for drug response prediction.
  Cong Shen, Guancen Lin, Chuan-Shen Hu, Jiawei Luo — Bioinformatics (Oxford, England), 2026-08-03
  MolDr models molecules as multiscale simplicial complexes and integrates cellular profiles for robust drug response prediction, highlighting the value of higher-order topology.
  Affiliations: State Key Laboratory of Mathematical Sciences, Academy of Mathematics and Systems Science; Department of Applied Mathematics, National University of Kaohsiung; ...

• 📰 DynaTCR: dynamic hard-negative ensemble graph learning improves TCR-epitope binding prediction.
  Xiangzheng Fu, Xinyu Zhang, Linlin Zhuo, Yifan Chen, Dongsheng Cao, Quan Zou — Bioinformatics (Oxford, England), 2026-08-03
  DynaTCR combines PLM embeddings, a variance-preserving graph encoder, and hard-negative mining to improve TCR–epitope binding prediction under strict evaluation.
  Affiliations: Faculty of Computer Science and Artificial Intelligence, Shenzhen University of Advanced Technology; School of Data Science and Artificial Intelligence, Wenzhou University of Technology; ...

• 📰 Evolutionary profiles for protein fitness prediction.
  Xiaoran Jiao, Shengdong Lin, Jigang Fan, Zhanming Liang, Weian Mao, Hao Chen, Chunhua Shen — Bioinformatics (Oxford, England), 2026-08-03
  EvoIF fuses homologous evolutionary profiles with inverse folding logits to deliver lightweight, robust protein fitness prediction competitive with much larger PLMs.
  Affiliations: Computer Science and Technology, Zhejiang University; School of Information Science and Engineering, East China University of Science and Technology; ...

• 💻 Image-guided Spatial Omics Enhancement reveals Hidden Spatial Microstructures.
  Jiahao Liu, Gongning Luo, Qiaoming Liu, Suyu Dong, Guohua Wang, Yuming Zhao — Bioinformatics, 2026-08-08
  Bell/mmBell fuse histology images, spatial coordinates, and omics with adaptive attention to enhance resolution and denoise spatial transcriptomics across platforms.

• 📰 An Artificial Intelligence (AI) model integrating multiscale foundation model histopathology representations with molecular and clinical features predicts early and late distant recurrence in TAILORx
  Sparano, Joseph A., Lama, Norsang, ..., Norman, Sledge, George W., Jr — npj Breast Cancer, 2026-08-05
  IICM+ integrates histopathology foundation-model features, transcriptomics, and clinicopathology to predict early/late distant recurrence in HR+/HER2− breast cancer.

• 📰 A Generalizable and Interpretable Framework for Molecular Subtype Classification of Pancreatic Ductal Adenocarcinoma Integrating Conformal Uncertainty Quantification and Consensus-Based Explainable Artificial Intelligence Across Multiple Cohorts
  Ş. Yaşar, F. H. Yagin, Sarah A. Alzakari, Amal K. Alkhalifa, Fahaid Al-Hashem, A. Tabnjh — International Journal of Molecular Sciences, 2026-08-04
  Builds a PDAC subtype classifier with conformal uncertainty and consensus XAI that generalizes across cohorts and maintains calibrated predictions under domain shifts.

🧬 Preprints (arXiv + bioRxiv)

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

• 🧬 Dual-Specific Antibody Design Using Artificial Intelligence
  Peer, M.; Amit, I.; Diesendruck, Y.; Erlich, Z.; ...; Hadar, D.; Voropaev, A.; Fastman, Y.; Ofran, Y. — bioRxiv, 2026-08-05
  Presents an AI-assisted platform that designs dual-specific “multibody” antibodies by optimizing two paratopes within a single Fv, delivering therapeutic-grade candidates with high affinity, specificity, and functional superiority. Two designs have advanced to IND-enabling studies, showcasing translational impact for antibody therapeutics.
  Affiliations: Biolojic Design

• 🧬 Seed-Guided De Novo Design Expands the Structural Diversity of Antitoxin Protein Binders
  Britton, D.; Ghose, D. A.; Halpin, J. C.; Birnbaum, F.; ...; Raval, S.; Papanastasiou, M.; Carr, S. A.; Keating, A. E. — bioRxiv, 2026-08-04
  Guides RFdiffusion binder generation with PDB-derived surface-complementary “seed” fragments to target extended, multi-site interfaces, greatly increasing structural diversity and target contacts. Experimental screens yield high-affinity RelE neutralizers with novel binding modes and reduced cross-reactivity, expanding de novo binder design capabilities.
  Affiliations: Massachusetts Institute of Technology

• 🧬 Evolution-inspired multi-objective Bayesian optimization for protein engineering
  Wen, K.; Wang, S.; Sun, Y.; Li, S.; Wang, M.; Liu, H.; Li, Q.; Zhu, J. — bioRxiv, 2026-08-06
  Introduces EvoMOBO, an evolution-inspired multi-objective Bayesian optimization framework that couples path-dependent sequence generation with global competition for protein engineering. Validated on enzymes, it achieves large gains (e.g., 17.5%→95% conversion) and identifies aggregation-resistant variants using mechanism-derived labels.
  Affiliations: Jilin University

• 🧬 Evolutionary Design of Membrane-Lytic Antimicrobial Peptides with Mixture of Experts
  Dong, R.; Song, C. — bioRxiv, 2026-08-04
  Develops AMPainterV2, a mixture-of-experts, imitation-learning framework that evolves membrane-lytic antimicrobial peptides via insertion/deletion/mutation operators. All top-ranked designs are active in vitro, with miniaturized peptides retaining or improving potency, advancing mechanism-driven AMP discovery.
  Affiliations: Peking University

• 🧬 CryoLigATE: enhancing the resolvability of cryo-EM maps in protein-ligand complexes using deep learning
  Haloi, N.; Howard, R. J.; Lindahl, E. — bioRxiv, 2026-08-05
  Trains a hybrid conv–transformer model (CryoLigATE) to locally enhance ligand densities in cryo-EM maps around binding pockets. The method markedly improves ligand resolvability and preserves experimental features, enabling more confident structure-based drug design.
  Affiliations: Stockholm University

• 🧬 Move BeTween modAlities (MBTA) employs flow matching to predict single cell data modalities
  Xu, B.; Zhang, Y.; Michor, F. — bioRxiv, 2026-08-09
  Introduces MBTA, a flow-matching framework that maintains modality-specific latent spaces and learns cross-modal flows for single-cell multi-omics translation. It outperforms shared-latent approaches, resolving structural mismatch and enabling robust cross-modal inference in cancer and developmental datasets.
  Affiliations: Dana-Farber Cancer Institute

• 📄 LLM-Guided Retrieval for Prediction of Molecular Perturbation Responses
  Betty Xiong, Jan-Christian Huetter, Gabriele Scalia, Tommaso Biancalani, Sepideh Maleki — arXiv, 2026-08-03
  Proposes LLM-Guided Retrieval to predict drug-induced transcriptomic responses by ranking and aggregating biologically related compounds profiled in the same cell line. The retrieval-centric approach improves correlation and directional accuracy, especially for unseen-cell-line generalization on large single-cell perturbation atlases.

• 🧬 RiboRep: Replicate-Aware Cross-Modal Transformers for Codon-Resolved Ribosome Density Prediction
  Kuo, A.; Yue, Z.; Ku, W.-S.; Chen, H. — bioRxiv, 2026-08-07
  RiboRep is a replicate-aware cross-modal transformer that integrates nucleotide sequences and reference ribosome occupancy via dual encoders and cross-attention to predict codon-level ribosome density. It improves translational-state modeling across species and lays groundwork for translation-aware molecular digital twins.
  Affiliations: Auburn University

• 🧬 Boltz-Perturb: Improving Diversity and Accuracy in Protein-Ligand Co-Folding through Training-Free Conditioning Perturbation
  Jung, H.; Lee, B.; Cheng, A. C. — bioRxiv, 2026-08-05
  Introduces training-free conditioning perturbations (TBP/TCP) to co-folding models that unlock latent pose diversity and improve protein–ligand binding-mode prediction. The approach raises top-20 oracle success rates 2.6–7.8× with substantially less compute than high-temperature diffusion variants.
  Affiliations: Merck & Co., Inc.

• 📄 Spatial proteomics guided by H&E-based AI reveals recurrence-risk niches in triple-negative breast cancer
  Yesung Cho, Ji Hwan Park, Chanil Kim, Hyewon Kim, ..., Eun-Young Kim, Dongmyung Shin, Jongbae Park, In-Gu Do — arXiv, 2026-08-04
  Combines outcome-trained H&E recurrence-risk heatmaps with mass-spectrometry spatial proteomics to pinpoint molecularly distinct niches in TNBC. A 13-protein composite derived from AI-guided regions enhances survival stratification and integrates with image-derived risk for improved prognostication.

• 📄 MolBioKG: Grounding Out-of-Graph Molecules in Biomedical Knowledge Graphs via Multi-Resolution Structural Anchoring
  Yiming Zhang, Hikaru Shindo, Shuan Chen, Kaushalya Madhawa, Jun Jin Choong, Yuna Oikawa, Takashi Fujiwara, Keisuke Ozawa — arXiv, 2026-08-07
  MolBioKG grounds out-of-graph molecules into biomedical knowledge graphs via multi-resolution structural anchoring and an LLM policy for adaptive traversal. It boosts multi-hop reasoning and out-of-graph generalization while maintaining traceable structural anchors and evidence.

• 📄 ED-DiT: Physics-Guided Diffusion Pretraining for Transferable Molecular Representations from Electron Density
  Liang Shuang, Haocheng Wang, Jiayi Song, Shuquan Ye, Ben Fei — arXiv, 2026-08-04
  ED-DiT pretrains a Diffusion Transformer on electron-density point clouds with electron-number constraints to learn transferable molecular representations. It substantially improves electronic-structure-related tasks under limited supervision, including orbital energies and density prediction.

• 🧬 AI-guided discovery of antimicrobial peptides for urinary tract infections leveraging a new catalogue of the human urinary microbiome
  Ke, S.; Zingl, F. G.; Wang, X.-W.; Hale, V. L.; Weiss, S. T.; Waldor, M.; Liu, Y.-Y. — bioRxiv, 2026-08-05
  Builds a large urinary microbiome gene/MAG catalog and uses machine learning to mine antimicrobial peptides active against UTI-associated strains. Two predicted AMPs are experimentally validated against uropathogenic E. coli, demonstrating microbiome-guided therapeutic discovery.
  Affiliations: Harvard Medical School

• 📄 Physics-Based Molecular Fingerprints from Spectral Graph Theory Provide Efficient Geometry-Aware Measures of Chemical Similarity
  Jacob W. Toney, Ayleen Y. Farnood, Samir Darouich, Heather J. Kulik — arXiv, 2026-08-05
  Introduces physics-inspired 3D spectral graph fingerprints derived from Laplacian eigenvalues of complete molecular graphs with interaction-weighted edges. The representation is efficient, interpretable, E(3)-invariant, and outperforms 2D descriptors in distinguishing stereoisomers/conformers for ML and similarity tasks.

• 🧬 Graph Machine Learning for Physiological Role Prediction in Protein Contact Networks: A Large-Scale Comparative Study on the Human Proteome
  Cervellini, M.; Martino, A. — bioRxiv, 2026-08-05
  Performs a large-scale comparison of graph ML on protein contact networks to predict enzymatic function and EC classes across the human proteome. Deep GNNs excel in multiclass EC prediction, while classical kernels remain strong for binary enzymatic classification, highlighting complementary strengths.
  Affiliations: Department of AI, Data and Decision Sciences

• 📄 CrystalGRPO: Target-Aligned and Coverage-Preserving Reinforcement Learning for Flow-Based Crystal Structure Prediction
  Kaixiang Su, Hongfei Xue, Qiang Zhu — arXiv, 2026-08-06
  Presents CrystalGRPO, a reinforcement-learning post-training framework for flow-based crystal structure prediction that balances target recovery and candidate coverage. It improves Top-1 and Top-20 recovery across datasets and backbones by jointly optimizing coordinates and lattice states.

• 📄 ED-CSP: Crystal Structure Prediction from Electron Diffraction
  Germain Poloudenny, Yaël Frégier, Arnaud Demortière — arXiv, 2026-08-06
  ED-CSP predicts periodic crystal structures directly from multi-view electron diffraction spot patterns, composition, and atom count using a relational encoder and periodic flow generator. Trained on a large simulated ED dataset, it surpasses PXRDGen and demonstrates genuine generative capability beyond retrieval.

• 📄 Geometry-Informed Parameter-Efficient Fine-Tuning of Pre-trained Molecular GNNs for Blood-Brain Barrier Permeability Prediction
  Marco Vieto Vega, Long D. Nguyen, Binh P. Nguyen — arXiv, 2026-08-04
  BBBP-GeoPEFT introduces geometry-informed, parameter-efficient fine-tuning for pretrained molecular GNNs using multi-cutoff distance graphs and line graphs. It injects 3D geometric cues with minimal tunable parameters to improve blood–brain barrier permeability prediction under scaffold and random splits.

• 📄 Bi-semantic Chemical Embedder for Joint Representation Learning of SMILES and Natural Language
  David Ming Segura, Jeremy Goumaz, Joshua W. Sin, Bojana Ranković, Philippe Schwaller — arXiv, 2026-08-04
  CheMatE learns bi-semantic embeddings that jointly align SMILES and domain text via continued MLM followed by contrastive learning on synthetic SMILES–text pairs. The unified representation transfers well across molecular property prediction and scientific language tasks.

• 📄 DASyR-LLM: Domain-Aware Symbolic Regression with LLMs for Kinetic Model Discovery
  Roberto Aliaga Medina, Paulina Quintanilla, Antonio del Rio Chanona — arXiv, 2026-08-05
  DASyR-LLM integrates an LLM into symbolic regression loops to critique and propose kinetic rate expressions, injecting domain knowledge into model discovery. It reduces iterations by 42–79% and often proposes the correct structure, accelerating interpretable kinetic modeling in chemical/bioprocess systems.