Biology subjects

Jakubek, Y. A.

Publications and source records attributed to Jakubek, Y. A..

4 recordsLinked to original sources

bioRxiv2025

Systematic contextual biases in SegmentNT relevant to all nucleotide transformer models

Recent advances in large language models (LLMs) have extended to genomic applications, yet model robustness relative to context is unclear. Here, we demonstrate two intrinsic biases (input sequence length and nucleotide position) affecting SegmentNT results, a model included with the Nucleotide Transformer that provides nucleotide-level predictions of biological features. We demonstrate that nucleotide position within the input sequence (beginning, middle, or end) alters the nature of SegmentNTs raw prediction probabilities, which can be standardized to improve prediction consistency. While longer input sequence length improves model performance, diminishing returns suggest a surprisingly small input length of [~]3,072 nucleotides might be sufficient for many applications. We further identify a 24-nucleotide periodic oscillation in SegmentNTs prediction probabilities, revealing an intrinsic bias potentially linked to the models training tokenization (6-mers) and architecture. We identify potential approaches to account for these biases and provide generalizable insights for utilizing nucleotide-resolution functional prediction models.

bioinformatics↗

bioRxiv2022

Mosaic chromosomal alterations in blood across ancestries via whole-genome sequencing

Mosaic mutations in blood are common with increasing age and are prognostic markers for cancer, cardiovascular dysfunction and other diseases. This group of acquired mutations include megabase-scale mosaic chromosomal alterations (mCAs). These large mutations have mainly been surveyed using SNP array data from individuals of European (EA) or Japanese genetic ancestry. To gain a better understanding of mCA rates and associated risk factors in genetically diverse populations, we surveyed whole genome sequencing data from 67,390 individuals, including 20,132 individuals of African ancestry (AA), and 7,608 of Hispanic ancestry (HA) with deep (30X) whole genome sequencing data from the NHLBI Trans Omics for Precision Medicine (TOPMed) program. We adapted an existing mCA calling algorithm for application to WGS data, and observed higher sensitivity with WGS data, compared with array-based data, in uncovering mCAs at low mutant cell fractions. As in previous reports, we observed a strong association with age and a non-uniform distribution of mCAs across the genome. The presence of autosomal (but not chromosome X) mCAs was associated with an increased risk of both lymphoid and myeloid malignancies. After adjusting for age, we found that individuals of European ancestry have the highest rates of autosomal mCAs, mirroring the higher rate of leukemia in this group. Our analysis also uncovered higher rates of chromosome X mCAs in AA and HA compared to EA, again after adjusting for age. Germline variants in ATM and MPL showed strong associations with mCAs in cis, including ancestry specific variants. And rare variant gene-burden analysis confirmed the association of putatively protein altering variants in ATM and MPL with mCAs in cis. Individual rare variants in DCPS, ADM17, PPP1R16B, and TET2 were all associated with autosomal mCAs and rare variants in OR4C16 were associated with chromosome X mCAs in females. There was significant enrichment of co-occurrence of CHIP mutations and mCAs both altering cancer associated genes TET2, DNMT3A, JAK2, CUX1, and TP53. Overall, our study demonstrates that rates of mCAs differ across populations and that rare inherited germline variants are strongly associated with mCAs across genetically diverse populations. These results strongly motivate further studies of mCAs in under-represented populations to better understand the causes and consequences of this class of somatic variation.

genomics↗

bioRxiv2022

Integrated Molecular Characterization of Intraductal Papillary Mucinous Neoplasms: An NCI Cancer Moonshot Precancer Atlas Pilot Project

Intraductal papillary mucinous neoplasms (IPMNs) are cystic precursor lesions to pancreatic ductal adenocarcinoma (PDAC). IPMNs undergo multistep progression from low grade (LG) to high grade (HG) dysplasia, culminating in invasive neoplasia. While patterns of IPMN progression have been analyzed using multi-region sequencing for somatic mutations, there is no integrated assessment of molecular events, including copy number alterations (CNAs) and transcriptomics changes, that accompany IPMN progression. We performed laser capture microdissection on surgically resected IPMNs of varying grades of histological dysplasia obtained from 24 patients (total of 74 independent histological lesions), followed by whole exome and whole transcriptome sequencing. Overall, HG IPMNs displayed a significantly greater aneuploidy score than LG lesions, with chromosome 1q amplification, in particular, being associated with HG progression and with cases that harbored cooccurring PDAC. Furthermore, the combined assessment of single nucleotide variants (SNVs) and CNAs identified both linear and branched evolutionary trajectories, underscoring the heterogeneity in the progression of LG lesions to HG and PDAC. At the transcriptome level, upregulation of MYC-regulated targets and downregulation of transcripts associated with the MHC class I antigen presentation machinery was a common feature of progression to HG. Taken together, this work emphasizes the role of 1q copy number amplification as a putative biomarker of high-risk IPMNs, underscores the importance of immune evasion even in non-invasive precursor lesions, and supports a previously underappreciated role of CNA-driven branching evolution as an avenue for IPMN progression. Our study provides important molecular context for risk stratification and cancer interception opportunities in IPMNs. SignificanceIntegrated molecular analysis of genomic and transcriptomic alterations in the multistep progression of intraductal papillary mucinous neoplasms (IPMNs), which are bona fide precursors of pancreatic cancer, identifies features associated with progression of low-risk lesions to high-risk lesions and cancer, which might enable patient stratification and cancer interception strategies.

cancer biology↗

bioRxiv2019

Pan cancer patterns of allelic imbalance from chromosomal aberrations in 33 tumor types

Somatic copy number alterations (SCNAs), including deletions and duplications, serve as hallmarks of tumorigenesis. SCNAs may span entire chromosomes and typically result in deviations from an expected one-to-one ratio of alleles at heterozygous loci, leading to allelic imbalance (AI). The Cancer Genome Atlas (TCGA) reports SCNAs identified using a circular binary segmentation (CBS) algorithm, providing segment mean copy number estimates from Affymetrix single-nucleotide polymorphism DNA microarray total (log R ratio) intensities, but not allele-specific (\"B allele\") intensities that inform of AI. Here we seek to provide a TCGA-wide description of AI in tumor genomes, including AI induced by SCNAs and copy-neutral loss-of-heterozygosity (cnLOH), using a powerful haplotype-based method applied to allele-specific intensities. We present AI summaries for all 33 tumor sites and propose an automated adjustment procedure to improve calibration of existing SCNA calls in TCGA for tumors with high levels of aneuploidy where baseline intensities were difficult to establish without annotation of AI. Overall, 94% of tumor samples exhibited AI. Recurrent events included deletions of 17p, 9q, 3p, amplifications of 8q, 1q, 7p as well as mixed event types on 8p and 13q. The AI-based approach identified frequent cnLOH on 17p across multiple tumor sites, with additional site-specific cnLOH patterns. Our findings support the exploration of additional methods for robust automated inference procedures and to aid empirical discoveries across TCGA.

genomics↗