Biology subjects

Rogic, S.

Publications and source records attributed to Rogic, S..

9 recordsLinked to original sources

bioRxiv2026

Application of large language models to the annotation of cell lines and mouse strains in genomics data

Accurate, consistent and comprehensive metadata are essential for the reuse of functional genomics data deposited in repositories such as the Gene Expression Omnibus (GEO), however, achieving this often requires careful manual curation that is time-consuming, costly and prone to errors. In this paper, we evaluate the performance of Large Language Models (LLMs), specifically OpenAIs GPT-4o, as an assistive tool for entity-to-ontology annotation of two commonly encountered descriptors in transcriptomic experiments - mouse strains and cell lines. Using over 9,000 manually curated experiments from the Gemma database and over 5,000 associated journal articles, we assess the models ability to identify relevant free-text entries and map them to appropriate ontology terms. Using zero-shot prompting and retrieval-augmented generation (RAG) to incorporate domain-specific ontology knowledge, GPT-4o correctly annotated 77% of mouse strain and 59% of cell line experiments, and uncovered manual curation errors in Gemma for over 200 experiments. GPT-4o substantially outperformed a regular expression-based string-matching method, which correctly annotated only 6% of mouse strain experiments due to low precision. Model errors often arose from typographical mistakes or inconsistent naming in the GEO record or publication, and resembled those made by human curators. Along with annotations, our approach requests that the model output supporting context and quotes from the sources. These were typically accurate and enabled rapid curator verification. These findings suggest that LLMs are not ready to fully replace manual curators, but can already effectively support them. A human-in-the-loop workflow, in which LLMs annotations are provided to human curators for validation, may improve the efficiency and quality of large-scale biomedical metadata curation.

bioinformatics↗

bioRxiv2026

A Cell-Based Assay to Categorize Variants of Human Succinate Semialdehyde Dehydrogenase Associated with Autism Spectrum Disorder

Many genes are implicated in autism spectrum disorder (ASD) as a result of sequencing the genomes of individuals with ASD, however in most cases it remains unclear which genes are playing causative roles. ALDH5A1, which encodes the enzyme succinate semialdehyde dehydrogenase (SSADH), an important regulator of GABA metabolism, is causative in the syndrome SSADH deficiency and is also implicated in ASD. However, it is unknown how variants found in ASD affect SSADH function. We developed a yeast growth assay that models SSADH deficiency to quantify the functional impact of seven ALDH5A1 variants found in ASD. In this assay, expression of human ALDH5A1 partially complemented the growth defect caused by deletion of the ALDH5A1 ortholog UGA2. Using growth rate measurements, we calculated functional scores for 27 variants divided into calibration and ASD test variant groups. Functional scores for benign and pathogenic calibration variants segregated accordingly, validating the assay, while ASD variants displayed a range of activities from complete loss to normal function. Comparisons with published enzymatic assays and computational predictions showed broad agreement, while also identifying some limitations of these approaches. SUMMARY STATEMENTWe introduce a simple yeast assay to measure how autism-linked gene variants in ALDH5A1, which encodes the enzyme succinate semialdehyde dehydrogenase, affect its function, helping clarifying its role in autism.

neuroscience↗

bioRxiv2025

Persistent hindrances to data re-use in single-cell genomics

We report on our experience attempting to re-use published and publicly available single-cell (or single-nucleus) RNA-sequencing studies (scRNA-seq) from the Gene Expression Omnibus (GEO). We screened GEO for human, mouse and rat scRNA-seq studies as potential candidates for inclusion in the Gemma database of re-annotated and re-analyzed transcriptome studies. Using semi-automated and manual curation, we assessed whether GEO datasets included cell-level expression count matrices and cell-type annotations. We found that there are steep challenges to data reuse. Only [~]40% of studies provided readily usable processed count data that could be reliably mapped to GEO metadata, and fewer than 10% included author-provided cell-type annotations. While raw sequencing data were available for the majority of studies, only a small proportion could be re-analyzed automatically without reliance on heuristics. Our findings show that existing practices for single-cell RNA-sequencing data distribution and sharing are insufficient for effective reuse, and highlight the urgent need for repositories to strengthen and enforce submission requirements, particularly for processed data and cell-type annotations.

genomics↗

bioRxiv2025

Using semantic search to find publicly available gene-expression datasets

Millions of high-throughput, molecular datasets have been shared in public repositories. have been shared in public repositories. Researchers can reuse such data to validate their own findings and explore novel questions. A frequent goal is to find multiple datasets that address similar research topics and to either combine them directly or integrate inferences from them. However, a major challenge is finding relevant datasets due to the vast number of candidates, inconsistencies in their descriptions, and a lack of semantic annotations. This challenge is first among the FAIR principles for scientific data. Here we focus on dataset discovery within Gene Expression Omnibus (GEO), a repository containing 100,000s of data series. GEO supports queries based on keywords, ontology terms, and other annotations. However, reviewing these results is time-consuming and tedious, and it often misses relevant datasets. We hypothesized that language models could address this problem by summarizing dataset descriptions as numeric representations (embeddings). Assuming a researcher has previously found some relevant datasets, we evaluated the potential to find additional relevant datasets. For six human medical conditions, we used 30 models to generate embeddings for datasets that human curators had previously associated with the conditions and identified other datasets with the most similar descriptions. This approach was often, but not always, more effective than GEOs search engine. Our top-performing models were trained on general corpora, used contrastive-learning strategies, and used relatively large embeddings. Our findings suggest that language models have the potential to improve dataset discovery, perhaps in combination with existing search tools.

genomics↗

bioRxiv2024

A multi-model approach defines function altering MECP2 missense variants identified in individuals with autism spectrum disorder

Pathogenic variants in MECP2 commonly lead to Rett syndrome, where MECP2s function as a DNA cytosine methylation reader is believed critical. MECP2 variants are also catalogued in individuals with autism spectrum disorder (ASD), including nine missense variants which had no known clinical significance at the start of this study. To assess these nine variants as risk alleles for ASD, we developed MECP2 variant functional assays using budding yeast and Drosophila. We calibrated these assays with known pathogenic and benign variants. Our data predict that four ASD variants are loss of function and five are functional. Protein destabilization offers insight into the altered function of some of these variants. Notably, yeast and Drosophila lack DNA methylation, yet all Rett pathogenic and ASD variants located in the methyl DNA binding domain that we analyzed proved to be loss of function, suggesting a clinically-relevant role for non-methyl DNA-binding by MECP2.

neuroscience↗

bioRxiv2021

ModelMatcher: A scientist-centric online platform to facilitate collaborations between stakeholders of rare and undiagnosed disease research

Next-generation sequencing is a prevalent diagnostic tool for undiagnosed diseases, and has played a significant role in rare disease gene discovery. While this technology resolves some cases, others are given a list of possibly damaging genetic variants necessitating functional studies. Productive collaborations between scientists, clinicians, and patients can help resolve such medical mysteries, and provide insights into in vivo function of human genes. Furthermore, facilitating interactions between scientists and research funders, including non-profit organizations or commercial entities, can dramatically reduce the time to translate discoveries from bench to bedside. Several systems designed to connect clinicians and researchers with a shared gene of interest have been successful. However, these platforms exclude some stakeholders based on their role or geography. Here we describe ModelMatcher, a global online matchmaking tool designed to facilitate cross-disciplinary collaborations, especially between scientists and other stakeholders of rare and undiagnosed disease research. ModelMatcher is integrated into the Rare Diseases Models and Mechanisms Network and Matchmaker Exchange, allowing users to identify potential collaborators in other registries. This living database decreases the time from when a scientist or clinician is making discoveries regarding their genes of interest, to when they identify collaborators and sponsors to facilitate translational and therapeutic research.

genetics↗

bioRxiv2020

Multi-parametric analysis of 58 SYNGAP1 variants reveal impacts on GTPase signaling, localization and protein stability

SYNGAP1 is a Ras and Rap GTPase with important roles in regulating excitatory synaptic plasticity. While many SYNGAP1 missense and nonsense mutations have been associated with intellectual disability, epilepsy, schizophrenia and autism spectrum disorder (ASD), there are many variants of unknown significance (VUS). In this report, we characterize 58 variants in nine assays that examine multiple aspects of SYNGAP1 function. Specifically, we used multiplex phospho-flow cytometry to measure the impact of variants on pERK, pGSK3{beta} and pCREB and high-content imaging to examine their subcellular localization. We find variants ranging from complete loss-of-function (LoF) to wildtype (WT)-like in their ability to regulate pERK and pGSK3{beta}, while all variants retain at least partial ability to regulate pCREB. Interestingly, our assays reveal that a high percentage of variants located within the disordered domain of unknown function that makes up the C-terminal half of SYNGAP1 exhibited LoF, compared to the more well studied catalytic domain. Moreover, we find protein instability to be a major contributor to dysfunction only for two missense variants both located within the catalytic domain. Using high-content imaging, we find variants with nuclear enrichment/exclusion and aberrant nuclear speckle localization. These variants are primarily located within the C2 domain known to mediate membrane lipid interactions. We find that mislocalization is distinct from altered catalytic activity, highlighting multiple independent molecular mechanisms underlying variant dysfunction. Our multidimensional dataset allows clustering of variants based on functional phenotypes and provides high-confidence pathogenicity classification.

genetics↗

bioRxiv2019

Multi-model functionalization of disease-associated PTEN missense mutations identifies multiple molecular mechanisms underlying protein dysfunction

Functional variomics provides the foundation for personalized medicine by linking genetic variation to disease expression, outcome and treatment, yet its utility is dependent on appropriate assays to evaluate mutation impact on protein function. To fully assess the effects of 106 missense and nonsense variants of PTEN associated with autism spectrum disorder, somatic cancer and PHTS, we take a deep phenotypic profiling approach using 18 assays in 5 model systems spanning diverse cellular environments ranging from molecular function to neuronal morphogenesis and behavior. Variants inducing instability occurred across the protein, resulting in partial to complete loss of function (LoF), which was well correlated across models. However, assays were selectively sensitive to variants located in substrate binding and catalytic domains, which exhibited complete LoF or dominant negativity independent of effects on stability. Our results indicate that full characterization of variant impact requires assays sensitive to instability and a range of protein functions.

genetics↗

bioRxiv2019

VariCarta: a comprehensive database of harmonized genomic variants found in ASD sequencing studies

BackgroundRecent years has seen a boom in the application of the next-generation sequencing technology to the study of human diseases, including Autism Spectrum Disorder (ASD), where the focus has been on identifying rare, possibly causative genomic variants in ASD individuals. Because of the high genetic heterogeneity of ASD, a large number of subjects is needed to establish evidence for a variant or gene ASD-association, thus aggregating data across cohorts and studies is necessary. However, methodological inconsistencies and subject overlap across studies complicate data aggregation.\n\nDescriptionHere we present VariCarta, a web-based database developed to address these challenges by collecting, reconciling and consistently cataloguing literature-derived genomic variants found in ASD subjects using ongoing semi-manual curation. The careful manual curation combined with a robust data import pipeline rectifies errors, converts variants into a standardized format, identifies and harmonizes cohort overlaps and documents data provenance. The harmonization aspect is especially important since it prevents the potential double-counting of variants which can lead to inflation of gene-based evidence for ASD-association.\n\nConclusionVariCarta is the largest collection of systematically curated, harmonized and comprehensively annotated literature-derived ASD-associated variants. The database currently contains 35,615 variant events from 8,044 subjects, collected across 50 publications, and reconciles 6,057 variants that have been reported in literature multiple times. VariCarta is freely accessible at http://varicarta.msl.ubc.ca.

genomics↗