Weekly BioML Digest [August 17, 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)

1322 matched filters -> 20 selected after LLM relevance + novelty ranking.

• 🛠️ Aligning protein-generative models to experimental fitness with ProteinDPO
  Widatalla, Talal, Borah, Ashir A., ..., Rafailov, Rafael, Hie, Brian L. — Nature Methods, 2026-08-14
  Aligns a structure‑conditioned protein language model to experimental stability via Direct Preference Optimization, enabling generation and scoring of thermostable proteins and stabilizing large multichain complexes (e.g., hemagglutinin).

• 📡 ESMDynamic: Fast and accurate prediction of protein dynamic contact maps from single sequences
  Kleiman, Diego E., Feng, Jiangyan, Xue, Zhengyuan, Shukla, Diwakar — Nature Communications, 2026-08-11
  Introduces ESMDynamic to predict residue–residue contact dynamics from sequence using an ESMFold‑based model, matching ensemble MD methods while orders faster and enabling proteome‑scale analysis of conformational variability.

• 🧫 Mechanistic machine learning for prediction of prime editing outcomes
  Hsu, Alvin, Chen, Peter J., ..., Osborn, Mark J., Liu, David R. — Nature Biotechnology, 2026-08-12
  Develops OptiPrime, a mechanistic ML model for prime editing efficiency across PE3/twinPE, learning repair determinants and nominating MMR‑evasive silent edits to improve therapeutic editing in primary human and mouse cells.

• 🖥️ Robust generative transition-state models for unseen chemistry
  Darouich, Samir, Toney, Jacob W., ..., Niepert, Mathias, Kulik, Heather J. — Nature Computational Science, 2026-08-12
  Presents self‑supervised pretraining for generative transition‑state prediction that generalizes to unseen chemistry, including transition metal reactions, improving TS geometry and barrier estimates with reduced fine‑tuning data.

• 🏛️ 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
  Trains a contrastive protein language model to co‑embed sequences and structures, unifying global protein similarity and outperforming state‑of‑the‑art PLMs on downstream classification while recapitulating ECOD/CATH hierarchies.
  Affiliations: Department of Computer Science, University of Haifa; Department of Biochemistry and Molecular Biology, School of Neurobiology; ...

• 🏛️ AI-driven PROTAC design overcomes oncogenic resilience by eliminating the CLIP1-LTK fusion protein.
  Shicheng Chen, Haiting Duan, Sheng Zhong, Jingxuan Ge, ..., Xiaowu Dong, Jinxin Che, Tingjun Hou, Peichen Pan — Proceedings of the National Academy of Sciences of the United States of America, 2026-08-11
  Combines deep learning–guided ternary complex design with structure optimization to create DCL05, an oral PROTAC that degrades the CLIP1‑LTK fusion protein (DC50 ~40 pM) and overcomes resistance beyond kinase inhibition.
  Affiliations: College of Pharmaceutical Sciences, Zhejiang University; Department of Neurosurgery, State Key Laboratory of Oncology in South China; ...

• 🛠️ Lessons learned from a Kaggle challenge for particle picking in cryo-electron tomography
  Peck, Ariana, Hutchings, Joshua, ..., Harrington, Kyle I. S., Paraan, Mohammadreza — Nature Methods, 2026-08-14
  Benchmarks >1,000 Kaggle submissions for cryo‑ET particle picking, yielding ML models that outperform state‑of‑the‑art tools and providing a curated, public reference for robust in situ macromolecular annotation.

• 📰 Histological aging signatures for monitoring tissue-specific aging and disease
  Abila, Ernesto, Buljan, Iva, ..., Schiller, Herbert B., Rendeiro, André F. — Nature Medicine, 2026-08-14
  Builds deep learning tissue ‘age clocks’ from 25,712 histology slides across 40 organs, linking morphological aging to telomeres, comorbidities, and blood transcriptomics for noninvasive organ‑specific aging surveillance.

• 📰 Deep learning-based quantification of collagen and associated features from H&E-stained whole slide pathology images across cancer types
  Nguyen, Tan H., Zhang, Jun, ..., Lee, Justin, Egger, Robert — Communications Medicine, 2026-08-14
  Trains a DL model (iQMAI) to infer collagen content and fiber morphology directly from H&E whole slides, revealing collagen–immune relationships and prognostic associations in pancreatic cancer.

• 📰 Deep Learning Predicts Mutations and Outcomes in Gastrointestinal Stromal Tumors.
  Arianna Bonetti, Van-Linh Le, Zunamys I Carrero, Fabian Wolf, ..., Antoine Italiano, Amandine Crombe, Jean-Michel Coindre, Jakob Nikolas Kather — Cancer research, 2026-08-14
  Applies deep learning to WSIs to predict KIT/PDGFRA mutations, TKI sensitivity, and recurrence‑free survival in 8,398 GISTs, achieving high AUCs and risk stratification comparable to established pathology scores.
  Affiliations: Else Kroener Fresenius Center for Digital Health, Faculty of Medicine and University Hospital Carl Gustav Carus; Department of Data and Digital Health, Institut Bergonié; ...

• 📰 Generalizable cancer detection from ultra-low-pass WGS via deep contextual modeling of cfDNA sequences
  Xu, Yang, Wang, Song, ..., Wu, Xue, Shao, Yang — Molecular Biomedicine, 2026-08-12
  Develops a transformer multiple‑instance framework (Fragmentia‑AI WGS) for ultra‑low‑pass cfDNA, enabling pan‑cancer detection with strong cross‑platform generalization and prognostic value for chemo‑immunotherapy.

• 📰 Can Multiomics Modeling Enable Accurate Prediction of Microsatellite Instability in Colorectal Cancer?
  W. Ao, Yijiang Huang, Guoqun Mao, Wei Wang, Danjiang Huang, Yongfei Zheng, Shuitang Deng — Academic radiology, 2026-08-13
  Integrates CT radiomics, histopathologic pathomics, and clinical data using deep learning to predict MSI in colorectal cancer with near‑perfect AUCs across internal/external validation cohorts.

• 📰 Label-free biochemical imaging and time point analysis of neural organoids via deep learning-enhanced Raman microspectroscopy.
  Dimitar Georgiev, Ruoxiao Xie, Daniel Reumann, Xiaoyu Zhao, Álvaro Fernández-Galiana, Mauricio Barahona, Molly M Stevens — Science advances, 2026-08-14
  Combines Raman microspectroscopy with deep learning–based hyperspectral unmixing to map biochemical changes in intact neural organoids over development, enabling label‑free 3D phenotyping.
  Affiliations: Department of Computing, and UKRI Centre for Doctoral Training in AI for Healthcare; Department of Materials, Department of Bioengineering; ...

• 📰 Pixel super-resolved fluorescence lifetime imaging using deep neural networks
  Casteleiro Costa, Paloma, Ghapandar Kashani, Parnian, ..., Marcu, Laura, Ozcan, Aydogan — PhotoniX, 2026-08-14
  Introduces a cGAN‑based pixel super‑resolution for FLIM (FLIM_PSR_k), enabling 5× resolution gain and faster acquisition on patient‑derived tissue with significant improvements in lifetime image quality.

• 📰 S2site: accurate protein binding site prediction with geometric deep learning and protein language model
  Liu, Zijing, Zhang, Linwei, Kong, Lupeng, Li, Yu — BMC Bioinformatics, 2026-08-13
  Proposes S2Site, a geometric deep learning + protein language model that predicts protein binding sites (PPI, antibody–antigen, peptide) at residue level, outperforming MSA‑dependent baselines.

• 📰 MiLaSol: Modeling Protein Solubility by Mixing Up Multiple Protein Language Models
  Weiwei Lou, Mert Erden, Lenore J. Cowen — Bioinformatics Advances, 2026-08-14
  MiLaSol ensembles multiple protein language models for solubility prediction (MCC 0.63) and couples simulated annealing to suggest solubilizing sequence edits validated across independent predictors.

• 📰 Sequence-structure cross-attention model integrating ESM-2 embeddings and AlphaFold cues for accurate prediction of Escherichia coli protein solubility.
  Z Elmi — Computers in biology and medicine, 2026-08-15
  SeqStruct‑XAttn fuses ESM‑2 embeddings with AlphaFold cues via cross‑attention for E. coli protein solubility, improving R2 to ~0.58 and transferring to yeast with calibrated outputs.
  Affiliations: Department of Software Engineering, Beykoz University

• 💻 Hierarchical Discrete Representations for Coarse-to-Fine Protein Conformation Generation.
  Seokjun On, Yujin Jeong, Kang-Hyeon Kim, Kyungheon Kang, Eun-Sol Kim — Bioinformatics, 2026-08-14
  See hierarchical discrete representation above; discrete, coarse‑to‑fine protein conformation generation improves over ESMDiff on dynamic ensembles.

• 📰 MMAllo: a multimodal deep learning framework and GaMD simulations for predicting protein allosteric sites and allosteric mutations
  Zhang, Pengyin, He, Yi, ..., Lai, Rui, Han, Weiwei — Journal of Cheminformatics, 2026-08-14
  MMAllo integrates PLM embeddings, graph structure, and ligand fingerprints to predict allosteric binding residues and mutation effects, recovering known pockets and validating a disease mutation via GaMD and MM/PBSA.

• 📰 EGA-DTA: An Energetic-Geometric Augmented Graph Neural Network With Target-Conditional Gating for DTA Prediction.
  Zhilin Zhu, Yifan Wu, Junkai Wang, G. Luo, Zhangli Lu, Min Li — IEEE transactions on computational biology and bioinformatics, 2026-08-14
  EGA‑DTA augments GNNs with energetic (BDE) and geometric (bond length) edge features plus protein‑conditional gating to improve drug–target affinity prediction, including cold‑start settings on KIBA/Davis/Metz.

🧬 Preprints (arXiv + bioRxiv)

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

• 🧬 SpatialAgent: An Autonomous AI Agent for Spatial Biology
  Wang, H.; He, Y.; Coelho, P. P.; Bucci, M.; ...; Copping, R.; Rozenblatt-Rosen, O.; Leskovec, J.; Regev, A. — bioRxiv, 2026-08-12
  Introduces an autonomous LLM-based agent for spatial biology that plans, executes, and verifies full analysis loops (panel design, annotation, trajectory, CCI), outperforming baselines and guiding a successful 100-gene Xenium add-on panel in mouse prostate cancer.
  Affiliations: Genentech, Stanford

• 🧬 PARNET: A CLIP-SEQ-BASED FOUNDATION MODEL FOR RNA SEQUENCE REPRESENTATION LEARNING
  Moyon, L.; Tirabassi, A.; Baranowskii, A.; Capitanchik, C.; ...; Gagneur, J.; Ule, J.; Horlacher, M.; Marsico, A. — bioRxiv, 2026-08-13
  Presents Parnet, a CLIP-seq–trained multi-task foundation model that predicts base-resolution RBP binding from sequence and yields general-purpose embeddings that match or exceed larger self-supervised RNA/genomic LMs across diverse downstream tasks with mechanistic interpretability.
  Affiliations: Computational Health Center, Helmholtz Center Munich; Cluster for Nucleic acid Sciences and Technologies - NUCLEATE, Munich

• 🧬 PerturbLDM: conditional latent diffusion for modelling single-cell perturbation responses
  Yu, L.; Hsieh, K.-L.; Chu, Y.; Lan, Q.; ...; Zhi, D.; Zhao, Z.; Jiang, X.; Dai, Y. — bioRxiv, 2026-08-16
  Develops a conditional latent diffusion model (PerturbLDM) to generate single-cell transcriptional responses to drugs/doses/cell lines, pretrained on Tahoe-100M and outperforming state-of-the-art across thousands of held-out perturbation-context combinations.
  Affiliations: University of Texas health science center at Houston

• 🧬 Physically Grounded Generative Modeling of All-Atom Biomolecular Dynamics
  Feng, B.; Zhang, J.; Zhang, X.; Cao, H.; Zhang, M.; Barth, P.; Liu, Z.; Li, Y. — bioRxiv, 2026-08-15
  BioKinema is a physically grounded spatiotemporal diffusion model that generates continuous-time all-atom biomolecular trajectories with correct thermodynamics/kinetics, revealing ligand-driven conformational changes and rare unbinding pathways at a fraction of MD cost.
  Affiliations: International Digital Economy Academy

• 🧬 MG2Act: A Mechanism-Inspired Sequential Attention Framework for Molecular Glue Degradation Prediction
  zhuang, z.; teng, d.; Xu, X.; Fang, S.; ...; Zhang, S.; Wang, X.; Zheng, M.; Qin, C. — bioRxiv, 2026-08-13
  MG2Act encodes the mechanism of molecular glues via sequential cross-attention (E3 engagement then substrate recruitment) to predict CRBN-mediated degradation, prospectively identifying nanomolar degraders including a non-classical IMiD-core CDK4 degrader.
  Affiliations: Key Laboratory of Marine Drugs, Chinese Ministry of Education

• 📄 ScreenShot: A Foundation Model for Few-Shot Combination Drug Screening
  Antoine de Mathelin, Christopher Tosh, Wesley Tansey — arXiv, 2026-08-12
  ScreenShot is a hierarchical transformer foundation model for few-shot combination drug screening that performs in-context prediction directly from functional measurements and drives active-learning designs matching uniform screens at one-third the budget.

• 📄 Stochastic Control Policies for Robust Molecular Transition Path Sampling
  Jingqian Liu, Yu-Hsiang Wang, Yanru Qu, Ge Liu — arXiv, 2026-08-13
  Proposes stochastic rollout-based control policies (FS-TPS, LaS-TPS) for transition path sampling that improve rare-event exploration and robustness over deterministic baselines across alanine dipeptide, chignolin, and BBL protein systems.

• 🧬 Virtual-cell verification enables self-auditing AI discovery for immune rejuvenation
  You, Y.; Fan, X.; Li, G.; Deng, W.; ...; Kong, J.; Chen, J.; Liu, X.; Tian, L. — bioRxiv, 2026-08-11
  Builds a self-auditing AI framework with phenotype and virtual-cell verifiers plus an Analyzer–Planner–Auditor agent to validate and revise objectives for immune rejuvenation, improving module-level reversal in PBMC screens and generalizing to independent assays.
  Affiliations: Guangzhou National Laboratory

• 📄 DegradeQuery: Counterfactual Tuple Pretraining for Context-Aware PROTAC Degradation Prediction
  Dong Xu, Zhangfan Yang, Jiantao Wu, Zexuan Zhu, Jianqiang Li, Junkai Ji — arXiv, 2026-08-11
  DegradeQuery converts label-missing molecule–target–E3 tuples into counterfactual pretraining signals for context-aware PROTAC degradation prediction, achieving state-of-the-art performance and complementing PLM features on PROTAC-8K.

• 🧬 A comprehensive phage-bacteria interaction atlas links phage lineage and capsule serotype to genome-guided machine learning prediction in Klebsiella pneumoniae
  Selvakumar, H.; Noonan, A. J. C.; Rotman, E.; Alayouni, M.; ...; Roux, S.; Mimee, M.; Arkin, A. P.; Mutalik, V. K. — bioRxiv, 2026-08-13
  Creates a species-wide K. pneumoniae phage–host atlas and a genome-guided ML predictor (AUROC 0.882, AUPR 0.765) that identifies capsule/LPS biosynthesis and defense features via SHAP, informing cocktail design with in vivo validation constraints.
  Affiliations: Environmental Genomics and Systems Biology, Lawrence Berkeley National Laboratory; Biological Systems and Engineering, Lawrence Berkeley Nati

• 🧬 Supervised Deep Learning for Efficient Cryo-EM Image Alignment in Drug Discovery with cryoPARES
  Sanchez-Garcia, R.; Berndt, A.; Apelbaum, A.; Reeks, J.; Williams, P. A.; Poelking, C.; Deane, C.; Saur, M. — bioRxiv, 2026-08-10
  cryoPARES leverages supervised DL trained on pre-aligned datasets to predict particle poses and prune particles, enabling near–real-time cryo-EM reconstructions and rapid ligand-bound complex determination in drug discovery pipelines.
  Affiliations: Astex Pharmaceuticals

• 🧬 Ab initio side-chain sampling with PUD+ enables high-fidelity protein dynamics across AI-driven and classical simulations
  Wu, D.; Wang, T. — bioRxiv, 2026-08-11
  Releases PUD+, a 40M-conformation ab initio dataset emphasizing side-chain coverage, enabling ML force fields and reparameterized classical force fields that better capture protein folding, IDP dynamics, and ligand-binding behavior.
  Affiliations: Tsinghua University

• 🧬 FrustrAI-Seq: Scaling Local Energetic Frustration to the Protein Sequence Space
  Leusch, J.-P.; Poley-Gil, M.; Fernandez-Martin, M.; Schlensok, J.; ...; Bordin, N.; Rost, B.; Parra, R. G.; Heinzinger, M. — bioRxiv, 2026-08-12
  FrustrAI-Seq predicts local energetic frustration directly from protein sequence using PLM embeddings, enabling proteome-scale frustration profiling without structures within minutes while retaining biological relevance.
  Affiliations: Helmholtz Munich

• 📄 Is Retrieval All You Need? Assessment and Emergence of Novelty in Protein Structure Generation
  Tongyue Xu, Yijie Zhang, Mutian He, Lingdong Shen, Zhihong Liu, Tianlei Ying, Cheng Tan — arXiv, 2026-08-11
  Introduces Domain Retrieval Rate (DRR) to assess novelty in protein backbone generators and presents RetFold, a zero-training retrieval-and-linker baseline, showing much generated structure can be assembled from known domains at far lower cost.

• 📄 Probing and steering biology across Boltz-1s trunk-diffusion boundary
  Piotr Jedryszek, Tongmeng Xie, Adam Winnifrith, Alexander Hasson, Weronika Ślesak, George Wicks, Toby Winnifrith, Oliver M. Crook — arXiv, 2026-08-11
  Dissects representations across the trunk–diffusion boundary of an AlphaFold3-class model (Boltz-1) via linear probes/SAEs and causal steering, finding geometry persists while sequence chemistry attenuates, and showing decodability need not imply causal control.

• 🧬 bgnorm: A Generative Statistical Framework for Background Correction, Normalisation, and Quality Control in Multiplex Spatial Proteomics
  Kharbanda, M.; Tubelleza, R.; Tan, Y.; Tan, C. W.; ...; Belz, G.; Kulasinghe, A.; Salim, A.; Bhuva, D. D. — bioRxiv, 2026-08-11
  bgnorm models multiplex spatial proteomics intensities with a generative mixture of background, non-specific binding, and signal to perform background correction, QC, and normalization, yielding top marker-positivity classification across platforms.
  Affiliations: The University of Queensland

• 📄 CytoBERT: A Foundation Model for Cytometry Data
  Syed Abdul Haseeb Qadri, Bjarne C. Hiller, Felix Blanke, Vanja Sophie Cangalovic, ..., Tom Siegl, Sebastian Bader, Thomas Kirste, Martin Becker — arXiv, 2026-08-14
  CytoBERT is an open-weight foundation model for single-cell cytometry with heterogeneous marker panels, pretrained on >50M cells to learn transferable inter-marker relationships and enable robust cross-dataset transfer learning.

• 🧬 Protein language models and the long tail of functional diversity
  Vinod, R.; Char, S.; Amini, A. P.; Crawford, L. K.; Yang, K. K. — bioRxiv, 2026-08-14
  Shows that singleton protein sequences—often excluded from PLM training—carry learnable signal and dense domain diversity, advocating their inclusion and dataset-specific clustering for better coverage of the long tail of protein function.
  Affiliations: Microsoft Research New England

• 🧬 LEN-Seek: Fast and scalable ligand binding-site similarity search in the latent space of an SE(3)-invariant graph VAE
  Yeo, K.; Kim, D.; Sim, J.; Lee, J. — bioRxiv, 2026-08-15
  LEN-Seek uses an SE(3)-invariant graph VAE with PLM-informed node features to embed ligand binding sites for fast latent-space similarity search, retrieving relevant templates at ~3,400× lower cost than geometric alignment.
  Affiliations: Department of Molecular Medicine and Biopharmaceutical Sciences, Graduate School of Convergence Science and Technology

• 🧬 Discovery of Selective Small-Molecule Ligands of SV2C by AI-Enhanced Virtual Screening and Experimental Validation
  Brueckner, A. C.; Martin, M. F.; Khuttan, S.; Shields, B.; ...; Salahpour, A.; Bucher, M. L.; Coleman, J. A.; Miller, G. W. — bioRxiv, 2026-08-16
  Combines homology modeling, MD/GaMD, and a CNN scoring function in an AI-enhanced virtual screening pipeline to discover selective SV2C ligands with micromolar affinity and isoform selectivity, validated experimentally.
  Affiliations: SandboxAQ, Palo Alto