Biology subjects

Cheng, A. C.

Publications and source records attributed to Cheng, A. C..

6 recordsLinked to original sources

bioRxiv2026

Model Validation Protocols for Machine Learning in Small Molecule Drug Discovery

Machine learning (ML) models for molecular property prediction are increasingly deployed in drug discovery, yet their adoption in real-world scenarios requires an understanding of the conditions in which a model succeeds or fails. While standardized benchmarks are powerful instruments to measure and unlock progress in ML research, they should not be blindly treated as the end goal. Especially static and retrospective benchmarks, in which no true unknown test set is employed, limit our ability to robustly validate a model's performance. Building on the collective expertise of a cross-industry consortium, we present a model validation framework consisting of five recommendations that would enable the community to move beyond aggregate metrics toward understanding where and why molecular property prediction models fail. We connect evaluation choices to real-world applications and case studies encountered in pharmaceutical research. The framework proposes splitting strategies that mimic realistic distribution shifts and expose common failure modes. We apply the recommended framework to a recently released dataset of absorption, distribution, metabolism, and excretion (ADME) properties. Across two complementary model algorithms, our case studies reveal four distinct failure modes (extrapolation, interpolation, representation, and evaluation), showing that model errors arise not only from distribution shift but also from limitations in molecular representations. Our results show that commonly used evaluation protocols can significantly overestimate performance and may not detect important model failure modes. All software and data are released via https://github.com/srijitseal/polaris.

bioinformatics↗

bioRxiv2026

Boltz-Perturb: Improving Diversity and Accuracy in Protein-Ligand Co-Folding through Training-Free Conditioning Perturbation

Protein-ligand co-folding models hold promise in structure-based drug discovery and small molecule interaction prediction, but often fail in predicting correct small molecule binding poses. We present Boltz-Perturb, a framework for addressing this through perturbing model conditioning signals during model inference, and show that such perturbations improve correct ligand binding mode predictions. We first show with true-coordinate injection experiments that the models learned energy landscape contains correct binding-mode basins, allowing us to reframe the problem as one of sampling deficiency. We then introduce two inference-time perturbation strategies, Token Bias Perturbation (TBP) and Token Conditioning Perturbation (TCP), which increase exploration of alternative binding poses. Across diverse protein-ligand systems, TCP improves top-20 oracle success rates by 2.6 to 7.8 fold. Boltz-Perturb attains higher oracle success rates compared to the Boltz-2 high diffusion temperature variant while requiring over 75% less compute. To our knowledge, this is the first systematic perturbation analysis of a co-folding architecture for small-molecule binding mode diversity. We demonstrate that inference-time perturbations can unlock latent structural diversity in generative co-folding models and improve protein-ligand predictions without costly retraining.

biophysics↗

bioRxiv2026

Defining influenza-specific B cells in vaccine responders, non-responders and influenza breakthrough infections

Although seasonal influenza vaccination programs are effective at a population level, our data from inactivated influenza vaccine (IIV) cohorts in years 2015-2022 reveal that 50-60% of individuals do not seroconvert following immunization. The underlying mechanisms of vaccine non-responsiveness are far from understood. In this study, we sought to define key determinants of optimal B cell immune responses elicited by seasonal influenza vaccination, and to explore why some individuals fail to elicit humoral immunity following immunization. Immune responses associated with seroconversion and vaccine failure from individuals immunized with IIVs were compared at cellular and molecular levels using single-cell transcriptomics. We analyzed HA-specific B cell immunity across vaccine-responders, breakthrough infections and patients hospitalized with acute influenza. Droplet-based single-cell RNA sequencing and VDJ-sequencing of influenza-specific B cells from stored PBMCs was performed using 10x Genomics. Our results show that atypical B cells are the major subset of B cell responses in vaccine non-responders on day 28 post-vaccination. Conversely, individuals who seroconvert had diverse B cell phenotypes. The use of recombinant influenza-specific HA probes allowed us to dissect expression patterns on influenza HA-specific B cells. We found that HA-specific B cells of vaccine non-responders for A/H1N1 and A/H3N2 components displayed elevated atypical-like markers (CD11c, FcRL-5) at baseline, compared to responders. Analysis of differentially expressed genes (DEGs) between responders and non-responders identified differential expression of HLA-DR, CD74, CD83, and CXCR3 genes. We subsequently demonstrated reduced frequencies of HLA-DR-, CD74- and CD83-expressing B cells in patients hospitalized with influenza, compared to healthy participants. Hospitalized influenza patients also had significantly higher proportions of atypical CD21-CD27- B cells. Overall, our data demonstrate an association between elevated frequencies of atypical-like B cells with both lack of seroconversion following immunization and severe influenza infection. These findings broaden our understanding of humoral immunity in influenza vaccination and infection, providing novel insights for vaccination strategies and design.

immunology↗

bioRxiv2024

Predicting functionally important breast cancer SNPs using pleiotropy, conservation, and protein structure

MotivationWith over 24,000 SNPs associated with breast cancer in ClinVar, there is a need to prioritize the subset most likely to be causally linked to diagnostic, prognostic, and other clinical outcomes of disease. Building off currently known breast cancer oncogenes and SNPs, we identify the subset of SNPs with pleiotropic effects, with the goal of identifying mutations functionally relevant to disease progression. We further use sequence and structure analysis to prioritize missense mutations most likely to impact protein function. ResultsFrom the known breast cancer SNPs, we identified co-associated mutations located at evolutionarily conserved positions and contributing significant protein stability as potential focal points for disease biomarkers, protein function studies, and therapeutic intervention. To identify regions likely integral to protein function, we plotted genomic intervals where multiple disease density peaks overlap. Of the breast cancer SNPs, 1,714 were co-associated in-frame mutations, of which 930 occurred at conserved residue positions (Shannon Entropy <1.0) and 833 were also missense mutations. Building structure-based models of the 277 SNPs with available protein structure resulted in identification of 133 SNPs that are calculated to affect protein thermostability by >100-fold (>3 kcal/mol). The workflow we built can be applied to other diseases to help identify functional mutations. AvailabilityPython code for the integrated analysis workflow available at http://github.com/mcarpenter-brandeis/brc-pleiotropy and detailed data tables available in Supplemental Information. Contactacheng@brandeis.edu, meredith.a.carpenter@gmail.coms

bioinformatics↗

bioRxiv2020

Discovery of A New Class of Integrin Antibodies for Fibrosis

Lung fibrosis, or the scarring of the lung, is a devastating disease with huge unmet medical need. There are limited treatment options and its prognosis is worse than most types of cancer. We previously discovered that MK-0429 is an equipotent pan-inhibitor of all v integrins that reduces proteinuria and kidney fibrosis in a preclinical model. In the present study, we further demonstrated that MK-0429 significantly inhibits fibrosis progression in a bleomycin-induced lung injury model. In search of newer integrin inhibitors for fibrosis, we characterized monoclonal antibodies discovered using Adimabs yeast display platform. We identified several potent neutralizing integrin antibodies with unique human and mouse cross-reactivity. Among these, Ab-31 blocked the binding of multiple v integrins to their ligands with IC50s comparable to those of MK-0429. Furthermore, both MK-0429 and Ab-31 suppressed integrin-mediated cell adhesion and latent TGF{beta} activation. In IPF patient lung fibroblasts, TGF{beta} treatment induced profound SMA expression in phenotypic imaging assays and Ab-31 demonstrated superior in vitro activity at inhibiting SMA expression, suggesting that the integrin antibody is able to modulate TGF{beta} action though mechanisms beyond the inhibition of latent TGF{beta} activation. Together, our results highlight the potential to develop newer integrin therapeutics for the treatment of fibrotic lung diseases. One Sentence Summarytargeting integrin in lung fibrosis

pharmacology and toxicology↗

bioRxiv2020

Capturing heterogeneous infectiousness in transmission dynamic models of tuberculosis: a compartmental modelling approach

Infectiousness heterogeneity among individuals with tuberculosis (TB) is substantial and is likely to have a significant impact on the long-term dynamics of TB and the effectiveness of interventions. However, there is a gap in capturing heterogeneous infectiousness and evaluating its impact on the effectiveness of interventions. Informed by observed distribution of secondary infections, we constructed a deterministic model of TB transmission using ordinary differential equations. The model incorporated assumption of heterogeneous infectiousness with three levels of infectivity, namely non-spreaders, low-spreaders and super-spreaders. We evaluated the effectiveness of dynamic transmission untargeted and targeted implementation of an intervention intended to represent active case finding with a point-of-care diagnostic tool. The simulated intervention detected 20% of all TB patients who would otherwise have been missed by the health system during their disease episode and was compared across four epidemiological scenarios. Our model suggested that targeting the active case finding intervention towards super-spreaders was more effective than untargeted intervention in all setting scenarios, with more effectiveness in settings with low case detection and high transmission intensity. For instance, a targeted intervention achieved a 42.2% reduction in TB incidence, while the untargeted intervention achieved only a 20.7% reduction over 20 years, given the same number of people treated. Although the most marked impact on equilibrium TB incidence came from the rate of late reactivation, the proportion of super-spreaders and their relative infectiousness had shown substantial impact. Targeting active case-finding interventions to highly infectious cases likely to be particularly beneficial in settings where case detection is poor. Heterogeneity-related parameters had an equivalent effect to several other parameters that have been established as being very important to TB transmission dynamics.

microbiology↗