Yachie, A. | Search bioRxiv

bioRxiv2024

Replicating a High-Impact Scientific Publication Using Systems of Large Language Models

Publications focused on scientific discoveries derived from analyzing large biological datasets typically follow the cycle of hypothesis generation, experimentation, and data interpretation. The reproduction of findings from such papers is crucial for confirming the validity of the scientific, statistical, and computational methods employed in the study, and it also facilitates the foundation for new research. By employing a multi-agent system composed of Large Language Models (LLMs), including both text and code generation agents built on OpenAIs platform, our study attempts to reproduce the methodology and findings of a high-impact publication that investigated the expression of viral-entry-associated genes using single-cell RNA sequencing (scRNA-seq). The LLM system was critically evaluated against the analysis results from the original study, highlighting the systems ability to perform simple statistical analysis tasks and literature reviews to establish the purpose of the analyses. However, we also identified significant challenges in the system, such as nondeterminism in code generation, difficulties in data procurement, and the limitations presented by context length and bias from the models inherent training data. By addressing these challenges and expanding on the systems capabilities, we intend to contribute to the goal of automating scientific research for efficiency, reproducibility, and transparency, and to drive the discussion on the role of AI in scientific discovery.

bioinformatics↗

bioRxiv2020

Artificial intelligence based computational framework for drug-target prioritization and inference of novel repositionable drugs for Alzheimer's disease

BackgroundIdentification of novel therapeutic targets is a key for successful drug development. However, the cost to experimentally identify therapeutic targets is huge and only 400 genes are targets for FDA-approved drugs. Therefore, it is inevitable to develop powerful computational tools to identify potential novel therapeutic targets. Because proteins make their functions together with their interacting partners, a protein-protein interaction network (PIN) in human could be a useful resource to build computational tools to investigate potential targets for therapeutic drugs. Network embedding methods, especially deep-learning based methods would be useful tools to extract an informative low-dimensional latent space that contains enough information required to fully represent original high-dimensional non-linear data of PINs. ResultsIn this study, we developed a deep learning based computational framework that extracts low-dimensional latent space embedded in high-dimensional data of the human PIN and uses the features in the latent space (latent features) to infer potential novel targets for therapeutic drugs. We examined the relationships between the latent features and the representative network metrics and found that the network metrics can explain a large number of the latent features, while several latent features do not correlate with all the network metrics. The results indicate that the features are likely to capture information that the representative network metrics can not capture, while the latent features also can capture information obtained from the network metrics. Our computational framework uses the latent features together with state-of-the-art machine learning techniques to infer potential drug target genes. We applied our computational framework to prioritized novel putative target genes for Alzheimers disease and successfully identified key genes for potential novel therapeutic targets (e.g., DLG4, EGFR, RAC1, SYK, PTK2B, SOCS1). Furthermore, based on these putative targets, we inferred repositionable candidate-compounds for the disease (e.g., Tamoxifen, Bosutinib, and Dasatinib) DiscussionsOur computational framework could be powerful computational tools to efficiently prioritize new therapeutic targets and drug repositioning. It is pertinent to note here that our computational platform is easily applicable to investigate novel potential targets and repositionable compounds for any diseases, especially for rare diseases.

bioinformatics↗