Biology subjects

Rubinsteyn, A.

Publications and source records attributed to Rubinsteyn, A..

16 recordsLinked to original sources

bioRxiv2026

Scientific computing in the age of agentic AI: an exploratory field report

Scientific computing has become a central component of modern scientific discovery. Yet many computational tools are developed by small, specialized teams under incentives that encourage the release of rapidly prototyped tooling without commensurate attention to engineering concerns, including performance and maintainability. These gaps are particularly visible in the life sciences, where the advent of high-throughput sequencing and molecular profiling has made the production and processing of datasets routine at scales that strain reliability and cost. Recently, LLM-based agents have become increasingly capable, with publicly available systems possessing both significant domain knowledge in many scientific fields and the ability to autonomously operate over complex and specialized codebases in pursuit of well-defined goals. Together, these developments create a practical opportunity for scientific computing. Many of the persistent weaknesses of the scientific computing ecosystem stem from technical debt and a shortage of sustained engineering labor and expertise. Here, we examine coding agents as a potential way to address these weaknesses: we present an exploratory field report of eight early case studies in the application of LLM agents to scientific computing across a range of project scopes, from lightweight maintenance tasks to full performance-oriented rewrites of scientific libraries, with a focus on the life sciences. Each of these case studies is accompanied by reflections from the individual or group responsible for the work, including lessons from the process. Overall, we find that the use of coding agents in scientific computing holds great promise for accelerating scientific research and increasing the reliability of critical systems, but that outstanding concerns remain, including responsibility and ownership for such projects, and we suggest collaboration and stewardship with existing maintainers when feasible.

bioinformatics↗

bioRxiv2026

Peptide:MHC Binding Stability Prediction Using Protein Language Models

AO_SCPLOWBSTRACTC_SCPLOWPeptide:MHC class I (pMHC-I) binding stability governs the persistence of antigenic complexes at the cell surface and plays a key role in facilitating downstream immunological signals such as antigen presentation, T-cell activation, and immunodominance. However, methods for in silico stability prediction remain underexplored relative to binding affinity prediction, in part because available half-life datasets are sparse and expensive to collect. Here, we perform a systematic reassessment of pMHC-I stability prediction using controlled, similarity-aware data splits and apply a recently introduced supervised transfer-learning strategy to MINT, an interaction-aware protein language model, pre-trained on binding affinity and fine-tuned for quantitative half-life prediction. We show that MINT improves stability prediction over standard ESM-2 representations and existing predictors, and that assay-conditioned recalibration corrects systematic shifts across experimental measurement modalities. Across eluted ligand, immunogenicity, and personalized neoantigen prioritization benchmarks, predicted stability provides signal beyond binding affinity, enriching for naturally presented and immunogenic peptides within affinity-filtered candidate sets. These results establish pMHC-I half-life as an orthogonal and transferable biophysical signal connecting peptide binding, surface presentation, and T-cell recognition, and provide a leakage-aware, assay-aware framework for future antigen-presentation modeling.

bioinformatics↗

bioRxiv2026

Long-read analysis of tetrameric microsatellites with vmwhere supports GGAA repeat length-dependent chromatin state association in Ewing sarcoma

Microsatellites are abundant genomic elements that contribute to genetic diversity and disease-associated regulatory variation. Although long-read sequencing enables accurate resolution of repetitive regions, computational methods for fully resolved microsatellite genotyping remain limited. Here, we introduce variant motif where (vmwhere), a computational framework for identifying, genotyping, decomposing, and visualizing complex tetrameric microsatellites from long-read sequencing data. Using simulated error-free reads, vmwhere accurately measures several genotyping metrics, including allele length, repeat length, maximum consecutive repeat length, and motif density. Applied to long-read whole-genome sequencing data, vmwhere identified sequence interruptions, motif-specific differences in repeat architecture, and ancestry-associated allele variation, including long repeat alleles that exceed short-read sequencing limitations. We applied vmwhere to GGAA microsatellites in Ewing sarcoma, an aggressive pediatric cancer driven by EWS-FLI1 fusion oncoprotein, which binds to microsatellites and remodels chromatin. Genome-wide integration of long-read-defined microsatellite architecture with chromatin accessibility and EWS-FLI1 binding revealed that GGAA repeat structure was associated with chromatin state, with longer consecutive repeat microsatellites exhibiting increased EWS-FLI1 binding and chromatin accessibility. Cell line-specific expansions and contractions of GGAA microsatellite repeat length were associated with gains and losses of chromatin accessibility. Further, we identified haplotype-specific chromatin states, with preferential binding and accessibility at longer alleles. Together, these results establish vmwhere as a scalable framework for resolving population-level microsatellite variation and linking repeat architecture to chromatin state. Repeat structure and length characteristics provides insights into genotype-function relationships at microsatellite repeats in cancer.

cancer biology↗

bioRxiv2025

PRAME Epitopes are T-Cell Immunovulnerabilities in BRD4::NUTM1 Initiated NUT Carcinoma

BackgroundNUT carcinoma is a rare but highly lethal solid tumor without an effective standard of care. NUT carcinoma is caused by bromodomain-containing NUTM1 fusion oncogenes, most commonly BRD4::NUTM1. BRD4::NUTM1 recruits p300 to acetylate H3K27 forming expansive stretches of hyperacetylated chromatin called "megadomains" with the overexpression of corresponding oncogenes, including MYC. We hypothesized that transcriptional dysregulation caused by BRD4::NUTM1 would lead to the generation of cancer-specific antigens that could be therapeutically actionable. MethodsWe integrated genomics, computational antigen prediction software, targeted immunopeptidomics using single- and double-labeled peptide standards, and gain/loss-of-function genetic experiments on a panel of cell lines (N=5), a patient derived xenograft, a tissue microarray (N=77), and patient samples from the Tempus AI Sequencing Database harboring evidence of NUTM1 fusions (N=165). We created an PRAME425 T-cell receptor x SP34 CD3 bispecific molecule modeled after brenetafusp, an PRAME425 T-cell receptor bispecific T-cell engager, as well as PRAME425 TCR T-cells based on anzutresgene autoleucel and we applied these products to NUT carcinoma cells in vitro. ResultsWe identified PRAME as the most commonly expressed cancer/testis antigen in patient samples harboring the three canonical NUT carcinoma fusions (BRD4::NUTM1, BRD3::NUTM1, and NSD3::NUTM1). Additionally, 56% (43/77) of NUT carcinoma tissue microarray samples stained positive for PRAME. BRD4::NUTM1 expression in HEK 293T cells enhanced PRAME levels and BRD4::NUTM1 knockout in NUT carcinoma cells reduced PRAME levels. Immunopeptidomics detected more PRAME-derived HLA ligands (N=9) than all other cancer/testis antigens combined (N=5). Targeted mass spectrometry detected the HLA-A*02:01/SLLQHLIGL (PRAME425) epitope in 100% (4/4) of HLA-A*02+, PRAME+ NUT carcinoma samples at higher levels (>0.01 fM) than HLA-A*02:01/RLDQLLRHV (PRAME312) or HLA-A*02:01/YLHARLREL (PRAME462). The PRAME425 T-cell receptor x SP34 CD3 bispecific molecule and PRAME425 TCR T-cells each exhibited potent, T-cell mediated cytotoxicity against PRAME+ NUT carcinoma cells. ConclusionsPRAME is highly and frequently expressed in NUT carcinoma and the most common oncoprotein causing NUT carcinoma, BRD4::NUTM1, contributes to these high PRAME levels. PRAME epitopes presented by HLA Class I are a previously unrecognized therapeutic vulnerability for NUT carcinoma that warrant clinical trials testing PRAME targeted immunotherapies in this neglected patient population. What is already known on this topicNUT carcinoma is a devastating malignancy that is recalcitrant to cytotoxic chemotherapy, T-cell checkpoint blockade, and targeted therapies in the form of bromodomain inhibitors. What this study addsNUT carcinoma tumors are high in the cancer/testis gene PRAME. The oncogene most commonly causing NUT carcinoma, BRD4::NUTM1, contributes to these high levels. NUT carcinoma cells present PRAME epitopes on HLA Class I molecules and are susceptible to PRAME-directed, T-cell mediated cytotoxicity. How this study might affect research, practice or policyOur results argue for phase I/II clinical trials testing PRAME immunotherapies like brenetafusp or anzutresgene autoleucel in PRAME+ NUT carcinoma patients. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=149 SRC="FIGDIR/small/642090v3_ufig1.gif" ALT="Figure 1"> View larger version (38K): org.highwire.dtl.DTLVardef@495654org.highwire.dtl.DTLVardef@c2b975org.highwire.dtl.DTLVardef@1de507org.highwire.dtl.DTLVardef@a76d33_HPS_FORMAT_FIGEXP M_FIG C_FIG

cancer biology↗

bioRxiv2024

TCR-TRANSLATE: Conditional Generation of Real Antigen Specific T-cell Receptor Sequences

The paradoxical nature of T-cell receptor (TCR) specificity, which requires both precise recognition and adequate coverage of antigenic peptide-MHCs (pMHCs), poses a fundamental challenge in immunology. Efforts at modeling this complex many-to-many mapping have been greatly impeded by a severe lack of experimental data. To address this, we present TCR-TRANSLATE, a novel framework that adapts low-resource machine translation techniques to the TCR:pMHC specificity domain. Here, we explore sequence-to-sequence (seq2seq) modeling with various training strategies, including semi-synthetic data augmentation and multi-task objectives to generate antigen specific TCR sequences for a given target of interest. We benchmark twelve model variants derived from the BART and T5 model architectures on a target-rich validation set of well-studied pMHCs, finding an optimal model, TCRT5, that generated validated antigen-specific CDR3{beta} sequences for previously unseen antigens. While current limitations include a narrow validation set and a focus on the CDR3{beta} loop, our approach demonstrates the potential of seq2seq models in rapidly generating antigen-specific TCR repertoires, offering a promising avenue for increasing throughput in precision immunotherapies. Our findings highlight both the capabilities and limitations of sequence-based conditional TCR design, emphasizing the need for experimental validation to bridge the gaps between predictions, metrics, and functional capacity.

synthetic biology↗

bioRxiv2024

Preliminary attempt to inhibit proliferation of NUT carcinoma cell lines using antisense oligonucleotides targeting NUTM1

NUT carcinoma is an aggressive cancer driven by NUTM1 fusion proteins. This study evaluated antisense oligonucleotides (ASOs) targeting NUTM1 as a potential therapeutic approach. Three ASOs targeting NUTM1 and scrambled controls were tested at 10-50 nM doses in three NUT carcinoma cell lines (TC-797, 10-15, 14169) and one control line (293T). ASOs were delivered both gymnotically and using a transfection reagent. Cell viability was assessed at 48 and 72 hours using a luminescence-based assay. No ASOs showed selective inhibition of NUT carcinoma cell viability compared to controls across all conditions tested. Thus, this pilot study did not identify ASOs with activity against NUT carcinoma cells. However, it did establish preliminary protocols and generated data to inform future ASO optimization efforts targeting NUTM1 in NUT carcinoma. Further refinement of ASO design and delivery methods is needed to identify therapeutic candidates. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=135 SRC="FIGDIR/small/605673v1_ufig1.gif" ALT="Figure 1"> View larger version (32K): org.highwire.dtl.DTLVardef@de8a08org.highwire.dtl.DTLVardef@383db9org.highwire.dtl.DTLVardef@5c4500org.highwire.dtl.DTLVardef@1b6325f_HPS_FORMAT_FIGEXP M_FIG C_FIG HypothesisThis experiment was performed as a pilot experiment to generate data to inform future work, which should include formal hypothesis testing with a reduced number of conditions. The hypothesis of this pilot experiment was that one or more of the non-scrambled antisense oligonucleotides (ASOs) targeting NUTM1 will selectively decrease proliferation and/or viability of NUT carcinoma cells, indicated by a reduction in luminescence (as compared to treated non-NUT carcinoma and untreated NUT carcinoma control cells) using the CellTiter-Glo Luminescent Cell Viability Assay.

cancer biology↗

bioRxiv2023

ACE Configurator for ELISpot (ACE): Optimizing Combinatorial Design of Pooled ELISpot Assays with an Epitope Similarity Model

The ELISpot assay is a powerful in vitro immunoassay that enables cost-effective quantification of antigen-specific T-cell reactivity. It is widely used in the context of cancer and infectious diseases to validate the immunogenicity of epitopes. While technological advances in hardware and software have kept pace with the need for increased throughput, assay design and deconvolution methodology have largely remained stagnant. Current methods for designing multiplexed ELISpot assays are restricted to preset configurations, lack support for high-throughput scenarios, and ignore peptide identity during pool assignment. We introduce the ACE Configurator for ELISpot (ACE) to address these gaps. ACE generates optimized peptide-pool assignments from highly customizable user inputs and handles positive peptide deconvolution using assay readouts. We present a novel sequence-aware pooling strategy, powered by a fine-tuned ESM-2 deep sequence model to identify immunologically similar peptides, reducing the number of false positives and subsequent confirmatory assays. To validate the performance of ACE using real-world datasets, we conducted a comprehensive benchmark study against various design heuristics, deconvolution methods, and experimental conditions, contextualizing design parameter choices with their impact on precision and number of total pools. Our results demonstrate ACEs capacity to further increase precision of identified immunogenic peptides, maximizing experimental efficiency at the bench-side. ACE is freely available as an executable with a graphical user interface and command-line interfaces at https://github.com/pirl-unc/ace.

bioinformatics↗

bioRxiv2022

Comparing Long Read Fusion Callers using Simulated Read Data

The advent of single-molecule third generation sequencing technologies provide new possibilities for the detection of fusion transcripts in sequencing data. Here, we test three long-read fusions detection tools on simulated data, compare various tooling parameters and compare the performance between long-read and short-read fusion detection tools. We also use our fusion transcript detection pipeline to describe fusions transcripts detected in U87 and U937 glioblastoma cell lines. We find that LongGF is the most capable of the long-read fusion detection tools at identifying the most simulated fusion transcripts. While the short read fusion transcript detection tool, Arriba, had similar recall to some of the long-read tools, its precision was found to be much lower. Several fusions with ample evidence were found in U87 and U937 cell lines.

bioinformatics↗

bioRxiv2022

LENS - Landscape of Effective Neoantigens Software

MotivationElimination of cancer cells by T cells is a critical mechanism of anti-tumor immunity and cancer immunotherapy response. T cells recognize cancer cells by engagement of T cell receptors with peptide epitopes presented by major histocompatibility complex (MHC) molecules on the cancer cell surface. Peptide epitopes can be derived from antigen proteins coded for by multiple genomic sources. Bioinformatics tools used to identify tumor-specific epitopes via analysis of DNA and RNA sequencing data have largely focused on epitopes derived from somatic variants, though a smaller number have evaluated potential antigens from other genomic sources. ResultsWe report here an open-source workflow utilizing the Nextflow DSL2 workflow manager, Landscape of Effective Neoantigen Software (LENS), which predicts tumor-specific and tumor-associated antigens from single nucleotide variants, insertions and deletions, fusion events, splice variants, cancer testis antigens, overexpressed self-antigens, viruses, and endogenous retroviruses. The primary advantage of LENS is that it expands the breadth of genomic sources of discoverable tumor antigens using genomics data. Other advantages include modularity, extensibility, ease of use, and harmonization of relative expression level and immunogenicity prediction across multiple genomic sources. We present an analysis of 115 acute myeloid leukemia (AML) samples to demonstrate the utility of LENS. We expect LENS will be a valuable platform and resource for T cell epitope discovery bioinformatics, especially in cancers with few somatic variants where tumor-specific epitopes from alternative genomic sources are an elevated priority. AvailabilityMore information about LENS, including workflow documentation and instructions, can be found at https://gitlab.com/landscape-of-effective-neoantigens-software Contactsteven_vensko@med.unc.edu, benjamin_vincent@med.unc.edu Supplementary informationSupplementary data are available at Bioinformatics online.

cancer biology↗

bioRxiv2022

SARS-CoV-2 peptide vaccine elicits T-cell responses in mice but does not protect against infection or disease

We vaccinated BALB/c mice with peptides derived from the SARS-CoV-2 proteome selected in silico to elicit T-cell responses and/or B-cell responses against linear epitopes. These peptides were administered in combination with either of two adjuvants, poly(I:C) and the STING agonist BI-1387466. Antibody responses against predicted linear epitopes were not observed but both adjuvants consistently elicited T-cell responses to the same peptides, which were primarily from the set chosen for predicted T-cell immunogenicity. The magnitude of T-cell responses was significantly higher with BI-1387466 compared with poly(I:C). Neither adjuvant group, however, provided any protection against infection with the murine adapted virus SARS-CoV-2-MA10 or from disease following infection. In light of more recent evidence for protection from severe disease mediated by CD8+ T-cells, we suspect that the epitopes selected for vaccination were not presented by infected murine cells.

immunology↗

bioRxiv2020

Landscape and Selection of Vaccine Epitopes in SARS-CoV-2

There is an urgent need for a vaccine with efficacy against SARS-CoV-2. We hypothesize that peptide vaccines containing epitope regions optimized for concurrent B cell, CD4+ T cell, and CD8+ T cell stimulation would drive both humoral and cellular immunity with high specificity, potentially avoiding undesired effects such as antibody-dependent enhancement (ADE). Additionally, such vaccines can be rapidly manufactured in a distributed manner. In this study, we combine computational prediction of T cell epitopes, recently published B cell epitope mapping studies, and epitope accessibility to select candidate peptide vaccines for SARS-CoV-2. We begin with an exploration of the space of possible T cell epitopes in SARS-CoV-2 with interrogation of predicted HLA-I and HLA-II ligands, overlap between predicted ligands, protein source, as well as concurrent human/murine coverage. Beyond MHC affinity, T cell vaccine candidates were further refined by predicted immunogenicity, viral source protein abundance, sequence conservation, coverage of high frequency HLA alleles and co-localization of CD4+ and CD8+ T cell epitopes. B cell epitope regions were chosen from linear epitope mapping studies of convalescent patient serum, followed by filtering to select regions with surface accessibility, high sequence conservation, spatial localization near functional domains of the spike glycoprotein, and avoidance of glycosylation sites. From 58 initial candidates, three B cell epitope regions were identified. By combining these B cell and T cell analyses, as well as a manufacturability heuristic, we propose a set of SARS-CoV-2 vaccine peptides for use in subsequent murine studies and clinical trials. O_FIG O_LINKSMALLFIG WIDTH=180 HEIGHT=200 SRC="FIGDIR/small/135004v1_ufig1.gif" ALT="Figure 1"> View larger version (40K): org.highwire.dtl.DTLVardef@1fc6599org.highwire.dtl.DTLVardef@1725a45org.highwire.dtl.DTLVardef@84a233org.highwire.dtl.DTLVardef@1b4c7aa_HPS_FORMAT_FIGEXP M_FIG C_FIG

immunology↗

bioRxiv2020

Improved predictive models of peptide presentation on MHC I

Computational prediction of the peptides presented on MHC class I proteins is an important tool for studying T cell immunity. The data available to develop such predictors has expanded with the use of mass spec to identify naturally-presented MHC ligands. In addition to elucidating binding motifs, the identified ligands also reflect the antigen processing steps that occur prior to MHC binding. Here, we developed an integrated predictor of MHC I presentation that combines new models for MHC I binding and antigen processing. Considering only peptides first predicted by the binding model to bind strongly to MHC, the antigen processing model is trained to discriminate published mass spec-identified MHC I ligands from unobserved peptides. The integrated model outperformed the two individual components as well as NetMHCpan 4.0 and MixMHCpred 2.0.2 on held-out mass spec experiments. Our predictors are implemented in the MHCflurry package, version 1.6.0 (github.com/openvax/mhcflurry).

immunology↗

bioRxiv2017

Computational pipeline for the PGV-001 neoantigen vaccine trial

This paper describes the sequencing protocol and computational pipeline for the PGV-001 personalized vaccine trial. PGV-001 is a therapeutic peptide vaccine targeting neoantigens identified from patient tumor samples. Peptides are selected by a computational pipeline which identifies mutations from tumor/normal exome sequencing and ranks mutant sequences by a combination of predicted Class I MHC affinity and abundance estimated from tumor RNA. The PGV pipeline is modular and consists of many independently usable tools and software libraries. We draw attention to three particular tools which may be useful to other groups working on neoantigen vaccination. (1) Epidisco is a workflow which orchestrates the parallel execution of the PGV pipeline, including both common steps such as alignment as well as tools which have been developed specifically for the PGV-001 trial. (2) Vaxrank uses somatic variants and tumor RNA reads to select peptides for inclusion in a patients vaccine. (3) Isovar is a library used by Vaxrank to determine the mutated protein sequence associated with a genomic variant from supporting tumor RNA reads. We hope that the functionality of these tools may extend beyond the specifics of the PGV-001 trial and enable other research groups in their own neoantigen investigations.

bioinformatics↗

bioRxiv2017

MHCflurry: open-source class I MHC binding affinity prediction

Machine learning prediction of the interaction between major histocompatibility complex I (MHC I) proteins and their small peptide ligands is important for vaccine design and other applications in adaptive immunity. We describe and benchmark a new open-source MHC I binding prediction package, MHCflurry. The software is a collection of allele-specific binding predictors incorporating a novel neural network architecture and adhering to software development best practices. MHCflurry outperformed the standard predictors NetMHC 4.0 and NetMHCpan 3.0 on a benchmark of mass spec-identified MHC ligands and showed competitive accuracy on a benchmark of affinity measurements. The accuracy improvement was due to substantially better prediction of non-9-mer peptide ligands, which offset a narrowly lower accuracy on 9-mers. MHCflurry was on average 8.6X faster than NetMHC and 44X faster than NetMHCpan; performance is further increased when a graphics processing unit (GPU) is available. MHCflurry is freely available to use, retrain, or extend, includes Python library and command line interfaces, and may be installed using standard package managers.

immunology↗

bioRxiv2017

Vaxrank: A Computational Tool For Designing Personalized Cancer Vaccines

Therapeutic vaccines targeting mutant tumor antigens (\"neoantigens\") are an increasingly popular form of personalized cancer immunotherapy. Vaxrank is a computational tool for selecting neoantigen vaccine peptides from tumor mutations, tumor RNA data, and patient HLA type. Vaxrank is freely available at www.github.com/openvax/vaxrank under the Apache 2.0 open source license and can also be installed from the Python Package Index.

bioinformatics↗

bioRxiv2016

Somatic Mutations and Neoepitope Homology in Melanomas Treated with CTLA-4 Blockade

Immune checkpoint inhibitors are promising treatments for patients with a variety of malignancies. Toward understanding the determinants of response to immune checkpoint inhibitors, it was previously demonstrated that somatic mutation burden is associated with benefit and a hypothesis was posited that neoantigen homology to pathogens may in part explain the link between somatic mutations and response. To further examine this hypothesis, we reanalyzed cancer exome data obtained from a previously published study of 64 melanoma patients treated with CTLA-4 blockade and a new dataset of RNA-Seq data from 24 of those patients. We found that the predictive accuracy does not increase as analysis narrows from somatic mutation burden to predicted MHC Class I neoantigens, expressed neoantigens, or homology to pathogens. Further, the association between somatic mutation burden and response is only found when examining samples obtained prior to treatment. Neoantigen and expressed neoantigen burden are also associated with response, but neither is more predictive than somatic mutation burden. Neither the previously-described tetrapeptide signature nor an updated method to evaluate neoepitope homology to pathogens were more predictive than mutation burden.

cancer biology↗