Mathew J. Koretsky

Known pathogenic variants (PVs) in Parkinson disease (PD) contribute to disease development but have yet to be fully explored by arrays on a large scale. This study evaluated genotyping success of the NeuroBooster array (NBA) and determined the frequencies of PVs across ancestries. We analyzed the presence and allele frequency of PVs in 28,710 PD cases, 9,614 other neurodegenerative disorder cases, and 15,821 controls across 11 ancestries within the Global Parkinson's Genetics Program (GP2) data set. Cluster plots were used to assess the quality of PVs genotyped on NBA. Genes previously predicted to have high or very high confidence of causing PD tend to have more PVs and are present across ancestry groups. Of 34 known PD gene PVs assessed, 25 were typed by NBA and classified as “good” (n = 12), “medium” (n = 4), or “bad” (n = 9) quality variants. Our results confirm the likelihood that established PD genes are pathogenic and highlight the importance of ancestrally diverse research in PD. We also show the usefulness of the NBA as a reliable tool for the genotyping of rare variants of PD.

SORL1, the gene encoding the SORLA protein, has arisen as a potential therapeutic target for Alzheimer’s disease (AD). Studies suggest that restoring SORLA function or its trafficking pathways, particularly the SORLA–retromer recycling system, may offer a promising strategy to slow or halt AD progression. While both rare and common SORL1 variants have been associated with increased AD risk, recent evidence suggests a potential involvement of SORL1 in other neurodegenerative conditions. This study assessed the contribution of SORL1 genetic variation to the risk of AD, related dementias (RD), and Parkinson’s disease (PD) using data from six large-scale biobanks, comprising 15,043 AD, 9,943 RD, and 42,763 PD cases, along with 111,969 controls across 11 ancestries. We identified 53 potentially disease-related SORL1 variants (CADD score > 20, MAC ≥ 2, annotated as protein-altering or splicing, and with the mutated allele present only in cases), including 41 novel and 12 previously reported variants. Three were found across multiple ancestries. Overall, 13 variants were found in AD-related cohorts, 5 in RD cohorts, and 35 in PD cohorts. Association analysis identified 10 nominally significant variants associated with AD and 5 with PD. The replication of multiple SORL1 variants across neurodegenerative diseases and ancestrally diverse populations underscores its potential broad genetic contribution to neurodegeneration and reinforces its relevance across distinct clinical phenotypes. Gene-based burden analysis did not reveal any significant cumulative effect of SORL1 variants in the populations tested. A family-based analysis identified a rare predicted-damaging variant in two East Asian families (11:121478242:G:A, p.R176Q) and two variants in two families of European ancestry (11:121514222:A:C, p.N371T; 11:121545392:G:A, p.V672M) that show some evidence of segregation in PD families. Although these variants were slightly more frequent in unrelated PD cases vs. controls, none of them showed statistically significant enrichment in PD, likely due to their very low frequency. Overall, our results extend the understanding of SORL1 beyond AD, suggesting a broader role in neurodegeneration and emphasizing the need for diverse population studies when evaluating genetic risk.

Common and rare variants in LRRK2 influence Parkinson’s disease (PD) risk across diverse populations. We investigated the p.A419V variant across multiple ancestry cohorts comprising over 200,000 individuals. In cases of East Asian ancestry, the variant was significantly associated with increased risk (OR = 2.9; 95% CI: 1.66–5.10; p = 0.0002), It lies on rare EAS haplotypes, and was not in linkage disequilibrium with other LRRK2 coding variants. Although not significant, meta-analysis of age of onset in EAS cases show a suggestive trend (β = –0.89 years; SE = 1.01; p = 0.380). LRRK2 protein modelling prediction indicated that binding sites for RAB8A, RAB10, RAB29 were in close proximity to the p.A419V variant within the ARM domain. Together, these findings confirm the p.A419V as a significant PD risk factor in EAS populations, as well as highlight disease-relevant variants in the ARM domain and the link with LRRK2-RAB signaling pathway.

Background: The genetic architecture of Parkinson’s disease (PD) varies considerably across ancestries, yet most genetic studies have focused on individuals of European descent, limiting insights into the genetic landscape of PD at a global scale. Methods: We conducted a large-scale, multi-ancestry investigation of causal and risk variants in PD-related genes. Using genetic datasets from the Global Parkinson’s Genetics Program (GP2), we analyzed sequencing and genotyping data from 105,588 individuals, including 63,837 affected and 41,751 unaffected, from eleven different ancestries. Approximately 29% of individuals included were from underrepresented populations. Findings: Our findings revealed shared and ancestry-specific patterns in the prevalence and variant spectrum. Overall, ∼2% of PD individuals carried a causative variant, with substantial variations across ancestries ranging from ∼0·5% in African to ∼7% in Middle Eastern and >10% Ashkenazi Jewish ancestries. Including disease-associated GBA1 and LRRK2 risk variants raised the yield to 13·6%, largely driven by GBA1 , except in East Asians, where LRRK2 risk variants dominated. GBA1 variants were most frequent globally, albeit with substantial differences in frequencies and variant spectra. While GBA1 variants were identified across all ancestries, frequencies ranged from ∼5% in Middle Eastern and South Asian to ∼52% in African ancestry. Similarly, LRRK2 variants showed ancestry-specific enrichment, with p.G2019S most frequently seen in Middle Eastern and Ashkenazi Jewish populations, and risk variants predominating in East Asians. Notably, clinical trials targeting specific genetic variants are currently primarily based in Europe and North America. Another globally PD-relevant gene was PRKN , with variant carriers identified across almost all ancestries. Interpretation: This large-scale, multi-ancestry assessment offers crucial insights into the population-specific genetic architecture of PD. It underscores the critical need for increased ancestral diversity in PD research to improve diagnostic accuracy, enhance our understanding of disease mechanisms across populations, and ensure equitable development and application of emerging precision therapies. Funding: Aligning Science Across Parkinson’s (ASAP) Global Parkinson’s Genetics Program (GP2)

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.

Elucidating the genetic contributions to Parkinson’s disease (PD) etiology across diverse ancestries is a critical priority for the development of targeted therapies in a global context. We conducted the largest sequencing characterization of potentially disease-causing, protein-altering and splicing mutations in 710 cases and 11,827 controls from genetically predicted African or African admixed ancestries. We explored copy number variants (CNVs) and runs of homozygosity (ROHs) in prioritized early onset and familial cases. Our study identified rare GBA1 coding variants to be the most frequent mutations among PD patients, with a frequency of 4% in our case cohort. Out of the 18 GBA1 variants identified, ten were previously classified as pathogenic or likely pathogenic, four were novel, and four were reported as of uncertain clinical significance. The most common known disease-associated GBA1 variants in the Ashkenazi Jewish and European populations, p.Asn409Ser, p.Leu483Pro, p.Thr408Met, and p.Glu365Lys, were not identified among the screened PD cases of African and African admixed ancestry. Similarly, the European and Asian LRRK2 disease-causing mutational spectrum, including LRRK2 p.Gly2019Ser and p.Gly2385Arg genetic risk factors, did not appear to play a major role in PD etiology among West African-ancestry populations. However, we found three heterozygous novel missense LRRK2 variants of uncertain significance overrepresented in cases, two of which — p.Glu268Ala and p.Arg1538Cys — had a higher prevalence in the African ancestry population reference datasets. Structural variant analyses revealed the presence of PRKN CNVs with a frequency of 0.7% in African and African admixed cases, with 66% of CNVs detected being compound heterozygous or homozygous in early-onset cases, providing further insights into the genetic underpinnings in early-onset juvenile PD in these populations. Novel genetic variation overrepresented in cases versus controls among screened genes warrants further replication and functional prioritization to unravel their pathogenic potential. Here, we created the most comprehensive genetic catalog of both known and novel coding and splicing variants potentially linked to PD etiology in an underserved population. Our study has the potential to guide the development of targeted therapies in the emerging era of precision medicine. By expanding genetics research to involve underrepresented populations, we hope that future PD treatments are not only effective but also inclusive, addressing the needs of diverse ancestral groups.

Background Known pathogenic variants in Parkinson’s disease (PD) contribute to disease development but have yet to be fully explored by arrays at scale. Objectives This study evaluated genotyping success of the NeuroBooster array (NBA) and determined the frequencies of pathogenic variants across ancestries. Method We analyzed the presence and allele frequency of 34 pathogenic variants in 28,710 PD cases, 9,614 other neurodegenerative disorder cases, and 15,821 controls across 11 ancestries within the Global Parkinson’s Genetics Program dataset. Of these, 25 were genotyped on NBA and cluster plots were used to assess their quality. Results Genes previously predicted to have high or very high confidence of causing PD tend to have more pathogenic variants and are present across ancestry groups. Twenty-five of the 34 pathogenic variants were typed by the NBA array and classified “good” (n=12), “medium” (n=4), and “bad” (n=9) variants. Conclusion Our results confirm the likelihood that established PD genes are pathogenic and highlight the importance of ancestrally diverse research in PD. We also show the usefulness of the NBA as a reliable tool for genotyping of rare variants for PD.

Copy Number Variations (CNVs) play pivotal roles in the etiology of complex diseases and are variable across diverse populations. Understanding the association between CNVs and disease susceptibility is significant in disease genetics research and often requires analysis of large sample sizes. One of the most cost-effective and scalable methods for detecting CNVs is based on normalized signal intensity values, such as Log R Ratio (LRR) and B Allele Frequency (BAF), from Illumina genotyping arrays. In this study, we present CNV-Finder, a novel pipeline integrating deep learning techniques on array data, specifically a Long Short-Term Memory (LSTM) network, to expedite the large-scale identification of CNVs within predefined genomic regions. This facilitates efficient prioritization of samples for time-consuming or costly subsequent analyses such as Multiplex Ligation-dependent Probe Amplification (MLPA), short-read, and long-read whole genome sequencing. We incorporate four genes to establish our methods—Parkin ( PRKN ), Leucine Rich Repeat And Ig Domain Containing 2 ( LINGO2 ), Microtubule Associated Protein Tau ( MAPT ), and alpha-Synuclein ( SNCA )—which may be relevant to neurological diseases such as Alzheimer’s disease (AD), Parkinson’s disease (PD), Progressive Supranuclear Palsy (PSP), or related disorders such as essential tremor (ET). By training our models on expert-annotated samples and validating them across diverse cohorts, including those from the Global Parkinson’s Genetics Program (GP2) and additional dementia-specific databases, we demonstrate the efficacy of CNV-Finder in accurately detecting deletions and duplications. Our pipeline outputs app-compatible files for visualization within CNV-Finder’s interactive web application. This interface enables researchers to review predictions and filter displayed samples by model prediction values, LRR range, and variant count in order to explore or confirm results. Our pipeline integrates this human feedback to enhance model performance and reduce false positive rates. Through a series of comprehensive analyses and validations using visual inspection, MLPA, short-read, and long-read sequencing data, we demonstrate the robustness and adaptability of CNV-Finder in identifying CNVs with regions of varied size, probe density, and noise. Our findings highlight the significance of contextual understanding and human expertise in enhancing the precision of CNV identification, particularly in complex genomic regions like 17q21.31. The CNV-Finder pipeline is a scalable, publicly available resource for the scientific community, available on GitHub ( https://github.com/GP2code/CNV-Finder ; DOI 10.5281/zenodo.14182563 ). CNV-Finder not only expedites accurate candidate identification but also significantly reduces the manual workload for researchers, enabling future targeted validation and downstream analyses in regions or phenotypes of interest.

In this paper, we leveraged Large Language Models(LLMs) to accelerate data wrangling and automate labor-intensive aspects of data discovery and harmonization. This work promotes interoperability standards and enhances data discovery, facilitating AI-readiness in biomedical science with the generation of Common Data Elements (CDEs) as key to harmonizing multiple datasets. Thirty-one studies, various ontologies, and medical coding systems served as source material to create CDEs from which available metadata and context was sent as an API request to 4th-generation OpenAI GPT models to populate each metadata field. A human-in-the-loop (HITL) approach was used to assess quality and accuracy of the generated CDEs. To regulate CDE generation, we employed ElasticSearch and HITL to avoid duplicate CDEs and instead, added them as potential aliases for existing CDEs. The generated CDEs are foundational to assess the interoperability potential of datasets by determining how many data set column headers can be correctly mapped to CDEs as well as quantifying compliance with permissible values and data types. Subject matter experts reviewed generated CDEs and determined that 94.0% of generated metadata fields did not require manual revisions. Data tables from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and the Global Parkinson’s Genetic Program (GP2) were used as test cases for interoperability assessments. Column headers from all test cases were successfully mapped to generated CDEs at a rate of 32.4% via elastic search.The interoperability score, a metric for dataset compatibility to CDEs and other connected datasets, based on relevant criteria such as data field completeness and compliance with common harmonization standards averaged 53.8 out of 100 for test cases. With this project, we aim to automate the most tedious aspects of data harmonization, enhancing efficiency and scalability in biomedical research while decreasing activation energy for federated research.

Genotyping single nucleotide polymorphisms (SNPs) is fundamental to disease research, as researchers seek to establish links between genetic variation and disease. Although significant advances in genome technology have been made with the development of bead-based SNP genotyping and Genome Studio software, some SNPs still fail to be genotyped, resulting in “no-calls” that impede downstream analyses. To recover these genotypes, we introduce Cluster Buster, a genotyping neural network and visual inspection system designed to improve the quality of neurodegenerative disease (NDD) research. Concordance analysis with whole genome sequencing (WGS) and imputed genotypes validated the reliability of predicted genotypes, with dozens of high-performing SNPs across LRRK2, APOE , and GBA loci achieving at least 90% concordance per SNP location. Further analysis of concordance between Genome Studio genotypes and imputed and WGS genotypes revealed discrepancies between the genotyping technologies, highlighting the need for selective application of Cluster Buster on SNP locations based on concordance rates. Cluster Buster’s implementation significantly reduces manual labor for recovering no-call SNPs, refining genotype quality for the Global Parkinson’s Genetics Program (GP2). This system facilitates better imputation and GWAS outcomes, ultimately contributing to a deeper understanding of genetic factors in NDDs.