Biology subjects

Bostrom, M.

Publications and source records attributed to Bostrom, M..

5 recordsLinked to original sources

bioRxiv2025

Evaluating Large Language Models for Gene-to-Phenotype Mapping: The Critical Role of Full-Text Database Access

Transformer-based large language models (LLMs) have demonstrated significant potential in the biological and medical fields due to their ability to effectively learn from large-scale, diverse datasets and perform a wide range of downstream tasks. However, LLMs are limited by issues such as information processing inaccuracies and data confabulation. These limitations hinder their utility for literature searches and other tasks requiring accurate and comprehensive extraction of information from extensive scientific literature. In this study, we evaluated the performance of various LLMs in accurately retrieving peer-reviewed literature and mapping correlations between 198 genes and six phenotypes: bone formation, cartilage formation, fibrosis, cell proliferation, tendon formation, and ligament formation. Our analysis included three types of models. First, standard transformer-based LLMs (ChatGPT-4o and Gemini 1.5 Pro). Second, specialized LLMs with dedicated custom databases containing peer-reviewed articles (SciSpace and ScholarAI). Third, specialized LLMs without dedicated databases (PubMedGPT and ScholarGPT). Using human-curated gene-to-phenotype mappings as the ground truth, we found that specialized LLMs with dedicated databases achieved the highest accuracy (>80%) in gene-to-phenotype mapping. Additionally, these models were able to provide relevant peer-reviewed publications supporting each gene-to-phenotype correlation. These findings underscore the importance of database augmentation and specialization in enhancing the reliability and utility of LLMs for biomedical research applications.

genetics↗

bioRxiv2025

Database-Augmented Transformer-Based Large Language Models Achieve High Accuracy in Mapping Gene-Phenotype Relationships

Transformer-based large language models (LLMs) have demonstrated significant potential in the biological and medical fields due to their ability to effectively learn from large-scale, diverse datasets and perform a wide range of downstream tasks. However, LLMs are limited by issues such as information processing inaccuracies and data confabulation, which hinder their utility for literature searches and other tasks requiring accurate and comprehensive extraction of information from extensive scientific literature. In this study, we evaluated the performance of various LLMs in accurately retrieving peer-reviewed literature and mapping correlations between 102 genes and four phenotypes: bone formation, cartilage formation, fibrosis, and cell proliferation. Our analysis included standard transformer-based LLMs (ChatGPT4o and Gemini1.5 Pro), fine-tuned LLMs with dedicated custom databases containing peer-reviewed articles (SciSpace and ScholarAI), and fine-tuned LLMs without dedicated databases (PubMedGPT and ScholarGPT). Using human-curated gene-to-phenotype mappings as the ground truth, we found that fine-tuned LLMs with dedicated databases (SciSpace and ScholarAI) achieved high accuracy (>80%) in gene-to-phenotype mapping. Additionally, these models were able to provide relevant peer-reviewed publications supporting each gene-to-phenotype correlation. These findings underscore the importance of database augmentation and finetuning in enhancing the reliability and utility of LLMs for biomedical research applications.

genomics↗

bioRxiv2023

Staphyloccocus aureus biofilm, in the absence of planktonic bacteria, produces factors that activate counterbalancing inflammatory and immune-suppressive genes in human monocytes

Staphyloccocus aureus (S. aureus) is a major bacterial pathogen in orthopedic periprosthetic joint infection (PJI). S. aureus forms biofilms that promote persistent infection by shielding bacteria from immune cells and inducing an antibiotic-resistant metabolic state. We developed an in vitro system to study S. aureus biofilm interactions with primary human monocytes in the absence of planktonic bacteria. In line with previous in vivo data, S. aureus biofilm induced expression of inflammatory genes such as TNF and IL1B, and their anti-inflammatory counter-regulator IL-10. S. aureus biofilm also activated expression of PD-1 ligands that suppress T cell function, and of IL-1RA that suppresses differentiation of protective Th17 cells. Gene induction did not require monocyte:biofilm contact and was mediated by a soluble factor(s) produced by biofilm-encased bacteria that was heat resistant and > 3 kD in size. Activation of suppressive genes by biofilm was sensitive to suppression by Jak inhibition. These results support an evolving paradigm that biofilm plays an active role in modulating immune responses, and suggest this occurs via production of a soluble vita-PAMP. Induction of T cell suppressive genes by S. aureus biofilm provides insights into mechanisms that suppress T cell immunity in PJI, and suggest that anti-PD-1 therapy that is modeled on immune checkpoint blockade for tumors may be beneficial in PJI.

immunology↗

bioRxiv2019

Intragenomic variability and extended sequence patterns in the mutational signature of ultraviolet light

Mutational signatures can reveal properties of underlying mutational processes and are important when assessing signals of selection in cancer. Here we describe the sequence characteristics of mutations induced by ultraviolet (UV) light, a major mutagen in several human cancers, in terms of extended (longer than trinucleotide) patterns as well as variability of the signature across chromatin states. Promoter regions display a distinct UV signature with reduced TCG>TTG transitions, and genome-wide mapping of UVB-induced DNA photoproducts (pyrimidine dimers) showed that this may be explained by decreased damage formation at hypomethylated promoter CpG sites. Further, an extended signature model encompassing additional information from longer patterns improves modeling of UV mutation rate, which may enhance discrimination between drivers and passenger events. Our study presents a refined picture of the UV signature and underscores that the characteristics of a single mutational process may vary across the genome.

genomics↗

bioRxiv2018

Elevated pyrimidine dimer formation at distinct genomic bases underlie promoter mutation hotspots in UV-exposed cancers

Sequencing of whole cancer genomes has revealed an abundance of recurrent mutations in gene-regulatory promoter regions, in particular in melanoma where strong mutation hotspots are observed adjacent to ETS-family transcription factor (TF) binding sites. While sometimes interpreted as functional driver events, these mutations have also been suggested to be due to locally inhibited DNA repair or, alternatively, locally increased propensity for UV damage. Here, we provide evidence that base-specific elevations in the efficacy of UV lesion formation underlie these mutations. First, we find that low-dose UV light induces mutations preferably at a known ETS promoter hotspot in cultured cells even in the absence of global or transcription-coupled nucleotide excision repair (NER), ruling out inhibited repair. Further, by genome-wide mapping of cyclobutane pyrimidine dimers (CPDs) shortly after UV exposure and thus before DNA repair, we find that ETS-related mutation hotspots exhibit a strong base-specific increase in CPD formation frequency. Analysis of a large whole genome cohort illustrates the widespread contribution of this effect to recurrent mutations in melanoma. While inhibited NER underlies a general increase in somatic mutation burden in regulatory regions, we conclude that the most recurrently mutated individual DNA bases arise instead due to locally favorable conditions for UV damage formation, thus explaining a key phenomenon in whole-genome cancer analyses.

cancer biology↗