Biology subjects

Kaminow, B.

Publications and source records attributed to Kaminow, B..

6 recordsLinked to original sources

bioRxiv2026

Physical priors improve performance of structure-based binding affinity models

Structure-based drug discovery is a widely used paradigm for the rational design of novel small molecule therapeutics. However, the benefits conferred by the use of structural information has seen limited adoption in machine learning, where ligand-only ("2D") models are still the industry standard for molecular property or binding affinity prediction. Structure-based ("3D") ML models for binding-affinity prediction promise to present a clear advantage, but have not yet overtaken existing 2D models. Here, we show that physics-based priors can improve predictive performance of structure-based models by comparing different model architectures with varying physical priors on several prediction tasks. We present the Modular Training and Evaluation of Neural Networks (mtenn) package, where we decompose affinity prediction into separate steps of embedding structure into learned representations and combining those embeddings into a predicted binding affinity. We consider both E(3)-invariant and E(3)-equivariant architectures to determine the importance of encoding roto-translational inductive biases, as well as different methods for combining learned embeddings. By first optimizing several aspects of model construction using the general purpose PDBBind dataset, we are able to improve the performance and data efficiency of structure-based models. When subsequently trained and evaluated on the COVID Moonshot small molecule drug discovery dataset, our tuned models perform on par with industry standard ligand-only models. Our decomposed model framework highlights that encoding some physical priors improves model performance, while more complex biases such as equivariance offer limited benefit. Additionally, structure-based models generalize better to an unseen target and display higher training efficiency. Overall, these results emphasize that structure-based models benefit from their ability to incorporate physics-informed constraints, giving promising directions for model architecture development. These results also suggest that the strength of these models may be in tasks specifically aimed at generalizability, providing guidelines for their use in early-stage drug discovery campaigns.

bioinformatics↗

bioRxiv2025

How many crystal structures do you need to trust your docking results?

Structure-based drug discovery relies on the prediction of protein-bound poses of new molecule designs, the accuracy of which can impact downstream prioritization. While it is expected that crystal structures of similar molecules would provide the best template for predicting the poses of new designs, the time and cost required motivates identifying a point of diminishing returns for collecting new structures. Using 403 crystal structures of SARS-CoV-2 main protease from the open science COVID Moonshot project, we explore the tradeoff between the cost and utility of obtaining crystal structures for accurately predicting poses of designed molecules. We observe that similar reference ligands enable superior pose prediction and show that success plateaus after approximately five crystal structures per generic Bemis-Murcko scaffold, exceeding 95% for the campaign's lead series. This work provides practical recommendations for resource allocation in structure-enabled drug discovery campaigns.

biophysics↗

bioRxiv2024

Lessons learned during the journey of data: from experiment to model for predicting kinase affinity, selectivity, polypharmacology, and resistance

Recent advances in machine learning (ML) are reshaping drug discovery. Structure-based ML methods use physically-inspired models to predict binding affinities from protein:ligand complexes. These methods promise to enable the integration of data for many related targets, which addresses issues related to data scarcity for single targets and could enable generalizable predictions for a broad range of targets, including mutants. In this work, we report our experiences in building KinoML, a novel framework for ML in target-based small molecule drug discovery with an emphasis on structure-enabled methods. KinoML focuses currently on kinases as the relative structural conservation of this protein superfamily, particularly in the kinase domain, means it is possible to leverage data from the entire superfamily to make structure-informed predictions about binding affinities, selectivities, and drug resistance. Some key lessons learned in building KinoML include: the importance of reproducible data collection and deposition, the harmonization of molecular data and featurization, and the choice of the right data format to ensure reusability and reproducibility of ML models. As a result, KinoML allows users to easily achieve three tasks: accessing and curating molecular data; featurizing this data with representations suitable for ML applications; and running reproducible ML experiments that require access to ligand, protein, and assay information to predict ligand affinity. Despite KinoML focusing on kinases, this framework can be applied to other proteins. The lessons reported here can help guide the development of platforms for structure-enabled ML in other areas of drug discovery.

biophysics↗

bioRxiv2021

STARsolo: accurate, fast and versatile mapping/quantification of single-cell and single-nucleus RNA-seq data

We present STARsolo, a comprehensive turnkey solution for quantifying gene expression in single-cell/nucleus RNA-seq data, built into RNA-seq aligner STAR. Using simulated data that closely resembles realistic scRNA-seq, we demonstrate that STARsolo is highly accurate and significantly outperforms pseudoalignment-to-transcriptome tools. STARsolo can replicate the results of, but is considerably faster than CellRanger, currently the most widely used tool for pre-processing scRNA-seq data. In addition to uniquely mapped reads, STARsolo takes account of multi-gene reads, necessary to detect certain classes of biologically important genes. It has a flexible cell barcode processing scheme, compatible with many established scRNA-seq protocols, and extendable to emerging technologies. STARsolo can quantify transcriptomic features beyond gene expression, which we illustrate by analyzing cell-type-specific alternative splicing in the Tabula Muris project.

bioinformatics↗

bioRxiv2020

Virtue as the mean: Pan-human consensus genome significantly improves the accuracy of RNA-seq analyses

The Human Reference Genome serves as the foundation for modern genomic analyses. However, in its present form, it does not adequately represent the vast genetic diversity of the human population. In this study, we explored the consensus genome as a potential successor of the current Reference genome and assessed its effect on the accuracy of RNA-seq read alignment. In order to find the best haploid genome representation, we constructed consensus genomes at the Pan-human, Super-population and Population levels, utilizing variant information from the 1000 Genomes project. Using personal haploid genomes as the ground truth, we compared mapping errors for real RNA-seq reads aligned to the consensus genomes versus the Reference genome. For reads overlapping homozygous variants, we found that the mapping error decreased by a factor of ~2-3 when the Reference was replaced with the Pan-human consensus genome. Interestingly, we also found that using more population-specific consensuses resulted in little to no increase over using the Pan-human consensus, suggesting a limit in the utility of incorporating more specific genomic variation. To assess the functional impact, we compared splice junction expression in the different genomes and found that the Pan-human consensus increases accuracy of splice junction quantification for hundreds of splice junctions.

bioinformatics↗

bioRxiv2019

High-throughput prediction of MHC Class I and Class II neoantigens with MHCnuggets

Computational prediction of binding between neoantigen peptides and major histocompatibility complex (MHC) proteins is an emerging biomarker for predicting patient response to cancer immunotherapy. Current neoantigen predictors focus on in silico estimation of MHC binding affinity and are limited by low positive predictive value for actual peptide presentation, inadequate support for rare MHC alleles and poor scalability to high-throughput data sets. To address these limitations, we developed MHCnuggets, a deep neural network method to predict peptide-MHC binding. MHCnuggets is the only method to handle binding prediction for common or rare alleles of MHC Class I or II, with a single neural network architecture. Using a long short-term memory network (LSTM), MHCnuggets accepts peptides of variable length and is capable of faster performance than other methods. When compared to methods that integrate binding affinity and HLAp data from mass spectrometry, MHCnuggets yields a fourfold increase in positive predictive value on independent MHC-bound peptide (HLAp) data. We applied MHCnuggets to 26 cancer types in TCGA, processing 26.3 million allele-peptide comparisons in under 2.3 hours, yielding 101,326 unique candidate immunogenic missense mutations (IMMs). Predicted-IMM hotspots occurred in 38 genes, including 24 driver genes. Predicted-IMM load was significantly associated with increased immune cell infiltration (p<2e-16) including CD8+ T cells. Notably, only 0.16% of predicted immunogenic missense mutations were observed in >2 patients, with 61.7% of these derived from driver mutations. Our results provide a new method for neoantigen prediction with high performance characteristics and demonstrate its utility in large data sets across human cancers.\n\nSynopsisWe developed a new in silico predictor of Major Histocompatibility Complex (MHC) ligand binding and demonstrated its utility to assess potential neoantigens and immunogenic missense mutations (IMMs) in 6613 TCGA patients.

immunology↗