omicSynth: An open multi-omic community resource for identifying druggable targets across neurodegenerative diseases

MDMNI-DGD: A novel graph neural network approach for druggable gene discovery based on the integration of multi-omics data and the multi-view network

Accurately predicting druggable genes is of paramount importance for enhancing the efficacy of targeted therapies, reducing drug-related toxicities and improving patients' survival rates. Nevertheless, accurately predicting candidate cancer-druggable genes remains a critical challenge in translational medicine due to the high heterogeneity and complexity of cancer data. In this study, we proposed a novel graph neural approach called Druggable Gene Discovery based on the Integration of Multi-omics Data and the Multi-view Network (MDMNI-DGD), aiming to predict and evaluate cancer-druggable genes. MDMNI-DGD integrated a comprehensive set of multi-omics data, including copy number variations, DNA methylation, somatic mutations, and gene expression profiles. Simultaneously, it constructed the multi-view gene association network based on protein-protein interactions (PPI), protein structural domains, gene co-expression, pathway co-occurrence, gene sequence and gene ontology. Compared to other state-of-the-art approaches, MDMNI-DGD exhibits excellent performance in key evaluation metrics such as AUROC and AUPR. Moreover, the case study has also demonstrated the efficacy of our approach in discovering potentially druggable genes. Among more than 20,000 protein-coding genes, MDMNI-DGD successfully identified 872 potentially druggable genes. The findings from this investigation may serve to bolster the assessment of pan-cancer druggable genes, potentially catalyzing the development of more personalized and efficacious therapeutic interventions.

CARDBiomedBench: A Benchmark for Evaluating Large Language Model Performance in Biomedical Research: A novel question-and-answer benchmark designed to assess Large Language Models' comprehension of biomedical research, piloted on Neurodegenerative Diseases

Backgrounds

Biomedical research requires sophisticated understanding and reasoning across multiple specializations. While large language models (LLMs) show promise in scientific applications, their capability to safely and accurately support complex biomedical research remains uncertain.

Methods

We present CARDBiomedBench, a novel question-and-answer benchmark for evaluating LLMs in biomedical research. For our pilot implementation, we focus on neurodegenerative diseases (NDDs), a domain requiring integration of genetic, molecular, and clinical knowledge. The benchmark combines expert-annotated question-answer (Q/A) pairs with semi-automated data augmentation, drawing from authoritative public resources including drug development data, genome-wide association studies (GWAS), and Summary-data based Mendelian Randomization (SMR) analyses. We evaluated seven private and open-source LLMs across ten biological categories and nine reasoning skills, using novel metrics to assess both response quality and safety.

Results

Our benchmark comprises over 68,000 Q/A pairs, enabling robust evaluation of LLM performance. Current state-of-the-art models show significant limitations: models like Claude-3.5-Sonnet demonstrates excessive caution (Response Quality Rate: 25% [95% CI: 25% ± 1], Safety Rate: 76% ± 1), while others like ChatGPT-4o exhibits both poor accuracy and unsafe behavior (Response Quality Rate: 37% ± 1, Safety Rate: 31% ± 1). These findings reveal fundamental gaps in LLMs' ability to handle complex biomedical information.

Conclusion

CARDBiomedBench establishes a rigorous standard for assessing LLM capabilities in biomedical research. Our pilot evaluation in the NDD domain reveals critical limitations in current models' ability to safely and accurately process complex scientific information. Future iterations will expand to other biomedical domains, supporting the development of more reliable AI systems for accelerating scientific discovery.

Multi-Omics Analysis for Identifying Cell-Type-Specific Druggable Targets in Alzheimer's Disease

Background

Analyzing disease-linked genetic variants via expression quantitative trait loci (eQTLs) is important for identifying potential disease-causing genes. Previous research prioritized genes by integrating Genome-Wide Association Study (GWAS) results with tissue-level eQTLs. Recent studies have explored brain cell type-specific eQTLs, but they lack a systematic analysis across various Alzheimer's disease (AD) GWAS datasets, nor did they compare effects between tissue and cell type levels or across different cell type-specific eQTL datasets. In this study, we integrated brain cell type-specific eQTL datasets with AD GWAS datasets to identify potential causal genes at the cell type level.

Methods

To prioritize disease-causing genes, we used Summary Data-Based Mendelian Randomization (SMR) and Bayesian Colocalization (COLOC) to integrate AD GWAS summary statistics with cell-type-specific eQTLs. Combining data from five AD GWAS, three single-cell eQTL datasets, and one bulk tissue eQTL meta-analysis, we identified and confirmed both novel and known candidate causal genes. We investigated gene regulation through enhancer activity using H3K27ac and ATAC-seq data, performed protein-protein interaction and pathway enrichment analyses, and conducted a drug/compound enrichment analysis with the Drug Signatures Database (DSigDB) to support drug repurposing for AD.

Results

We identified 27 candidate causal genes for AD using cell type-specific eQTL datasets, with the highest numbers in microglia, followed by excitatory neurons, astrocytes, inhibitory neurons, oligodendrocytes, and oligodendrocyte precursor cells (OPCs). PABPC1 emerged as a novel astrocyte-specific gene. Our analysis revealed protein-protein interaction (PPI) networks for these causal genes in microglia and astrocytes. We found the "regulation of aspartic-type peptidase activity" pathway being the most enriched among all the causal genes. AD-risk variants associated with candidate causal gene PABPC1 is located near or within enhancers only active in astrocytes. We classified the genes into three drug tiers and identified druggable interactions, with imatinib mesylate emerging as a key candidate. A drug-target gene network was created to explore potential drug targets for AD.

Conclusions

We systematically prioritized AD candidate causal genes based on cell type-specific molecular evidence. The integrative approach enhances our understanding of molecular mechanisms of AD-related genetic variants and facilitates the interpretation of AD GWAS results.

Prioritizing Parkinson's disease risk genes in genome-wide association loci

Recent advancements in Parkinson's disease (PD) drug development have been significantly driven by genetic research. Importantly, drugs supported by genetic evidence are more likely to be approved. While genome-wide association studies (GWAS) are a powerful tool to nominate genomic regions associated with certain traits or diseases, pinpointing the causal biologically relevant gene is often challenging. Our aim was to prioritize genes underlying PD GWAS signals. The polygenic priority score (PoPS) is a similarity-based gene prioritization method that integrates genome-wide information from MAGMA gene-level association tests and more than 57,000 gene-level features, including gene expression, biological pathways, and protein-protein interactions. We applied PoPS to data from the largest published PD GWAS in East Asian- and European-ancestries. We identified 120 independent associations with P < 5×10-8 and prioritized 46 PD genes across these loci based on their PoPS scores, distance to the GWAS signal, and presence of non-synonymous variants in the credible set. Alongside well-established PD genes (e.g., TMEM175 and VPS13C), some of which are targeted in ongoing clinical trials (i.e., SNCA, LRRK2, and GBA1), we prioritized genes with a plausible mechanistic link to PD pathogenesis (e.g., RIT2, BAG3, and SCARB2). Many of these genes hold potential for drug repurposing or novel therapeutic developments for PD (i.e., FYN, DYRK1A, NOD2, CTSB, SV2C, and ITPKB). Additionally, we prioritized potentially druggable genes that are relatively unexplored in PD (XPO1, PIK3CA, EP300, MAP4K4, CAMK2D, NCOR1, and WDR43). We prioritized a high-confidence list of genes with strong links to PD pathogenesis that may represent our next-best candidates for disease-modifying therapeutics. We hope our findings stimulate further investigations and preclinical work to facilitate PD drug development programs.

Identification of 27 allele-specific regulatory variants in Parkinson's disease using a massively parallel reporter assay

Genome wide association studies (GWAS) have identified a number of genomic loci that are associated with Parkinson's disease (PD) risk. However, the majority of these variants lie in non-coding regions, and thus the mechanisms by which they influence disease development, and/or potential subtypes, remain largely elusive. To address this, we used a massively parallel reporter assay (MPRA) to screen the regulatory function of 5254 variants that have a known or putative connection to PD. We identified 138 loci with enhancer activity, of which 27 exhibited allele-specific regulatory activity in HEK293 cells. The identified regulatory variant(s) typically did not match the original tag variant within the PD associated locus, supporting the need for deeper exploration of these loci. The existence of allele specific transcriptional impacts within HEK293 cells, confirms that at least a subset of the PD associated regions mark functional gene regulatory elements. Future functional studies that confirm the putative targets of the empirically verified regulatory variants will be crucial for gaining a greater understanding of how gene regulatory network(s) modulate PD risk.