Robert P. Sheridan

Matched molecular pairs (MMP) is an approach to finding local changes in structure that give rise to consistent changes in activity. Assuming molecules A and B are in a matched pair, one can imagine a “transformation”, i.e., adding, removing, or exchanging a small set of atoms that changes A to B. Typical machine-learning algorithms are meant to handle individual molecules A, and model the activity difference between A and the average activity, and are agnostic to the presence of similar compounds in the training set. In contrast, MMP looks for differences between similar compounds A and B and is agnostic to the absolute value of the activities of A and B. Machine learning models tend to be poor at predicting activity cliffs, while MMP is designed to identify them. Previously, we (Sheridan et al., J. Chem. Inf. Model. 2006, 46, 180–192) suggested a workflow T-ANALYZE that organizes and displays sets of closely related drug-sized compounds such that any consistent MMP-based structure-activity in a data set can be easily seen. Nowadays, peptides are under consideration as candidate drugs and the MMP approach should be extended to them. Peptides are small enough that they can be treated as connection tables, i.e., atoms connected with bonds. Alternatively, they can be written as sequences, i.e., a one-dimensional list of monomer names. In this paper, we adapt the T-ANALYZE workflow to peptides represented both as connection tables and as sequences. We demonstrate this on the PAMPA activity of the CycPeptMPDB dataset.

HELM is a hierarchical notation system for biopolymers that is an increasingly popular choice for representing peptides. In this system, each monomer name must uniquely identify a single monomer, and until now, chemists have named peptide monomers by hand using a limited set of rules. As the chemical space of monomers is practically unlimited, this naming process is too time consuming to be done by chemists, and is prone to inconsistency. To address this bottleneck, we developed two complementary automated naming approaches for HELM name generation: a zone-based namer which relies on hand-curated dictionaries of substructures, and a generative AI namer which relies on a pre-trained Large Language Model (LLM) with chemistry world knowledge. We evaluated both approaches on a set of publicly disclosed alpha amino acids. For a time-split test set, the zone-based namer produced usable names for 84% of cases: 54% matched the human-generated name exactly, and an additional 30% did not match, but were interpretable as the correct structure. The LLM-based namer produced usable names for 42% of cases: 33% of the names matched the human-generated name exactly, and an additional 9% were non-identical but still interpretable as the correct structure. In another 49% of cases, the outputs were considered close matches, and while not directly usable, they can still inform the generation of correct names. The zone-based namer is deterministic, fast, and independent of any previous names, explicitly handling zone order and chirality. The LLM-based namer can generalize from human‑named examples to propose names for novel chemical groups, helping fill library gaps. Together, the approaches offer complementary coverage: rules‑based precision and speed paired with data‑driven flexibility for novel moieties. Combining the strengths of both approaches paves the way for an automated peptide monomer naming process that can accelerate research by: (i) helping the community establish standardized guidelines for the rapidly expanding non-canonical amino acid (ncAA) space; (ii) streamlining integration with synthesis planning, inventory management, and analysis pipelines (iii) enabling more consistent annotation in ELNs, databases, and publications; and (iv) facilitating cross-lab replication and data exchange. Together, these advances can expedite method development, reduce ambiguity in collaboration, and ultimately support the discovery of novel peptides with improved drug-like properties.

Heterobifunctional degraders, a class of targeted protein degraders (TPDs), often occupy beyond-rule-of-five (bRo5) chemical space, where traditional passive permeability models─calibrated on drug-like molecules or peptides and based on topological descriptors─have limited applicability. This study explores the predictive value of three-dimensional (3D) descriptors for modeling passive membrane permeability of heterobifunctional degraders in bRo5 chemical space. These include intramolecular hydrogen bonds (IMHBs), radius of gyration (R_gyr), and 3D polar surface area (3D-PSA) derived from conformational ensembles generated from well-tempered metadynamics (WT-MetaD) in explicit chloroform. These 3D ensembles were further refined and Boltzmann-weighted using ANI-2x neural network potentials to better represent molecular flexibility and solvent-relevant low-energy conformers. Three machine learning regression models─random forest regression (RF), partial least-squares (PLS), and linear support vector machines (LSVM)─were trained and evaluated using two-dimensional (2D), 3D, and combined 2D + 3D molecular descriptor sets. The inclusion of 3D descriptors consistently improved predictive performance across all models, notably in the challenging bRo5 chemical space. In the best-performing case (PLS), cross-validated r² improved from 0.29 to 0.48 with the addition of 3D features. This highlights the added predictive power of 3D features for modeling passive permeability in the challenging bRo5 chemical space, where 2D descriptors alone underperform. Feature importance analysis identified the 3D descriptor R_gyr as the dominant contributor to passive permeability, with additional contributions from 3D-PSA and intramolecular hydrogen bonds (IMHBs). Specifically, on the same held-out split, the linear relationship between permeability and R_gyr appears stronger than the averaged model results (means over 100 randomized 50/50 splits), underscoring the role of R_gyr as a dominant predictor in this data set while highlighting the consistent gains achieved by incorporating 3D information beyond 2D descriptors. Together, these 3D features reflect molecular compactness, spatial polarity, and internal hydrogen bonding─key determinants of passive permeability. These results highlight the utility of physically meaningful, ensemble-derived 3D descriptors for improving permeability prediction and guiding the rational design of permeable compounds in bRo5 chemical space. The Amber-based molecular dynamics workflow developed in this study is broadly applicable to heterobifunctional degraders, facilitating the evaluation of permeability properties across diverse protein targets and E3 ligases.

As with other pharma companies, we maintain production QSAR models of ADMET end points and update them regularly. Here, for six ADMET end points, we examine the predictions of test set molecules on multiple versions of random forest models spanning a period of 10 years. For any given end point, the predictions for the majority of molecules are similar for all model versions. However, for a small minority of molecules, the prediction shifts substantially over the span of a few versions. For most molecules that shift, the prediction becomes more accurate at later times. This Perspective investigates metrics that can help indicate which molecules will shift substantially in prediction and when the shift will occur.

Interpreting genomic variants in tumor samples presents a challenge in research and the clinical setting. A major barrier is that information about variants is fragmented across disparate databases, and aggregating information from these requires building extensive infrastructure. To this end, we have developed Genome Nexus, a one stop shop for variant annotation, equipped with a powerful API for bulk annotation of variants and a user friendly interface for cancer researchers. Genome Nexus is available at https://www.genomenexus.org. It a) aggregates variant information from a large number of sources that are relevant to cancer research and clinical applications; b) allows high-performance programmatic access to the aggregated data via a unified API; c) provides a search interface and a reference page for individual cancer variants; d) provides user-friendly tools for annotating variants in patients; e) is freely available under an open source license and can be installed in a private cloud or local environment. Genome Nexus contains annotations from more than a dozen resources, including those that provide variant effect information (VEP), protein sequence annotation (Uniprot, Pfam, dbPTM), functional consequence prediction (Polyphen-2, Mutation Assessor, SIFT), population prevalence (gnomAD, dbSNP, ExAC), cancer population prevalence (Cancer Hotspots, SignalDB) and clinical actionability (OncoKB, CIViC, Clinvar). The annotations can be accessed through the website, the API, and a command line client. Genome Nexus is unique in providing a user friendly interface specific to cancer that allows high performance annotation of any variant. It is the main annotation service for the popular cancer genomics tool cBioPortal, which serves thousands of users daily. It is also offered as a standalone tool for annotation, allowing researchers and clinicians as well as genomic infrastructure developers to leverage it directly in their own workflows. For example, a local installation of Genome Nexus is used for annotating all variants in AACR Project GENIE. Citation Format: Ino de Bruijn, Xiang Li, Onur Sumer, Benjamin Gross, Robert Sheridan, Angelica Ochoa, Manda Wilson, Avery Wang, Hongxin Zhang, Aaron Lisman, Adam Abeshouse, Sander Rodenburg, Sjoerd van Hagen, Remond Fijneman, Gerrit Meijer, Nikolaus Schultz, Jianjiong Gao. Genome Nexus: A comprehensive resource for the annotation and interpretation of genomic variants in cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1156.

CBioPortal for Cancer Genomics is an open-source platform for interactive, exploratory analysis of large-scale cancer genomics data sets. cBioPortal provides a user-friendly interface that integrates genomic and clinical data, and provides a suite of visualizations and analyses, including OncoPrints, mutation “lollipop” plots, variant interpretation, group comparison, survival analysis, expression correlation analysis, alteration enrichment analysis, cohort and patient-level visualization. cBioPortal also integrates external tools including CIViC, Cancer Digital Slide Archive, Next-Generation Clustered Heat Map, IGV and Bioconductor to facilitate interpretation. The public site (https://www.cbioportal.org) is accessed by ~35,000 unique visitors each month and hosts data from >325 studies spanning individual labs and large consortia. In addition, >67 instances of cBioPortal are installed at academic institutions and pharmaceutical/biotechnology companies worldwide. In 2021 we added data from 32 studies, totaling >24,000 samples, to the public site. All data is also available in the cBioPortal Datahub: https://github.com/cBioPortal/datahub/. We also host a dedicated instance for AACR Project GENIE, enabling access to the GENIE cohort of >135,000 clinically sequenced samples from 19 institutions (https://genie.cbioportal.org). In addition, the GENIE Biopharma Collaborative (BPC) enables the collection of comprehensive clinical annotations, including response, outcome, and treatment histories. The first BPC release contains data from >1,800 non-small cell lung cancer samples and will be released in early 2022. The growing GENIE cohort and the BPC clinical data have driven a number of recent developments, including performance improvements (the load time for the GENIE cohort was reduced from minutes to seconds). To leverage the BPC clinical data, we enabled sample selection based on treatment status, extended support for outcome analysis, and enhanced the patient timeline representation to incorporate response data. Additional development work has focused on improvements to variant interpretation, enhancements to the Mutations tab, and support for novel molecular assays via the ‘generic assay’ data type. Documentation on these new features and many others is available at https://www.cbioportal.org/news. cBioPortal is fully open source (https://github.com/cBioPortal/) under a GNU Affero GPL license. Development is a collaborative effort among groups at Memorial Sloan Kettering Cancer Center, Dana-Farber Cancer Institute, Children’s Hospital of Philadelphia, Princess Margaret Cancer Centre, Bilkent University and The Hyve. We welcome open source contributions from others in the cancer research community. Citation Format: Jianjiong Gao, Tali Mazor, Ino de Bruijn, Adam Abeshouse, Diana Baiceanu, Ziya Erkoc, Elena Garcia Lara, Benjamin Gross, David M. Higgins, Prasanna K. Jagannathan, Priti Kumari, Ritika Kundra, Xiang Li, James Lindsay, Aaron Lisman, Divya Madala, Ramyasree Madupuri, Angelica Ochoa, Oleguer Plantalech, Sander Rodenburg, Baby A. Satravada, Robert Sheridan, Lucas Sikina, Jessica Singh, S. Onur Sumer, Yichao Sun, Pim van Nierop, Avery Wang, Manda Wilson, Hongxin Zhang, Gaofei Zhao, Sjoerd van Hagen, Kees van Bochove, Ugur Dogrusoz, Allison Heath, Adam Resnick, Trevor J. Pugh, Chris Sander, Ethan Cerami, Nikolaus Schultz. cBioPortal for cancer genomics [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1155.

In the pharmaceutical industry, where it is common to generate many QSAR models with large numbers of molecules and descriptors, the best QSAR methods are those that can generate the most accurate predictions but that are also insensitive to hyperparameters and are computationally efficient. Here we compare Light Gradient Boosting Machine (LightGBM) to random forest, single-task deep neural nets, and Extreme Gradient Boosting (XGBoost) on 30 in-house data sets. While any boosting algorithm has many adjustable hyperparameters, we can define a set of standard hyperparameters at which LightGBM makes predictions about as accurate as single-task deep neural nets, but is a factor of 1000-fold faster than random forest and ~4-fold faster than XGBoost in terms of total computational time for the largest models. Another very useful feature of LightGBM is that it includes a native method for estimating prediction intervals.

While Gaussian process models are typically restricted to smaller data sets, we propose a variation which extends its applicability to the larger data sets common in the industrial drug discovery space, making it relatively novel in the quantitative structure–activity relationship (QSAR) field. By incorporating locality-sensitive hashing for fast nearest neighbor searches, the nearest neighbor Gaussian process model makes predictions with time complexity that is sub-linear with the sample size. The model can be efficiently built, permitting rapid updates to prevent degradation as new data is collected. Given its small number of hyperparameters, it is robust against overfitting and generalizes about as well as other common QSAR models. Like the usual Gaussian process model, it natively produces principled and well-calibrated uncertainty estimates on its predictions. We compare this new model with implementations of random forest, light gradient boosting, and k-nearest neighbors to highlight these promising advantages. The code for the nearest neighbor Gaussian process is available at https://github.com/Merck/nngp.

Correction for ‘QSAR without borders’ by Eugene N. Muratov et al., Chem. Soc. Rev., 2020, DOI: 10.1039/d0cs00098a.

Word cloud summary of diverse topics associated with QSAR modeling that are discussed in this review.