Biology subjects

Galvin, B.

Publications and source records attributed to Galvin, B..

4 recordsLinked to original sources

bioRxiv2025

DINOcell: Learning Generalizable Perturbation Effects through Self-Distillation

AO_SCPLOWBSTRACTC_SCPLOWPredicting cellular responses to therapeutics is a promising approach for novel target discovery. However, state-of-the-art computational models designed to predict perturbation effects struggle to generalize and outperform simple baselines. We present DINOcell, a weakly supervised framework that adapts self-distillation to single-cell transcriptomics for predicting perturbation effects. We demonstrate that DINOcell outperforms baselines in predicting the effects of single gene perturbations. Furthermore, DINOcell accurately predicts non-additive effects of combination perturbations, indicating its capacity to model complex genetic interactions. Finally, we show that DINOcell learns representations that capture biological signals and is a promising, generalizable approach for in silico perturbation modeling, providing a valuable tool for accelerating therapeutic target discovery.

bioinformatics↗

bioRxiv2022

A hierarchical genotyping framework using DNA melting temperatures applied to adenovirus species typing

Known genetic variation for a pathogen, in conjunction with post-PCR melting curve analysis, can be leveraged to provide increased taxonomic detail for pathogen identification in commercial molecular diagnostic tests. The increased taxonomic detail may be used by clinicians and public health decision makers to observe circulation patterns, monitor for outbreaks, and inform local testing practices. We propose a method for expanding the taxonomic resolution of PCR diagnostic systems by incorporating a priori knowledge of the assay design and publicly available sequence information into a genotyping classification model. For multiplexed PCR systems, this framework is generalized to incorporate information from multiple assays that react with different gene targets of the same pathogen to increase classification accuracy. To illustrate the method, a hierarchical classification model is developed for the BioFire(R) Respiratory 2.1 Panel and the BioFire(R) Respiratory 2 Panel (collectively the BioFire Respiratory Panels - highly multiplexed PCR diagnostic tests) to predict the species of human adenovirus (HAdV) from Adenovirus Detected test results. Performance of the classification model was characterized via a 10-fold cross-validation on a labeled dataset and exhibited 95% {+/-} 4% accuracy. The model was then applied to the BioFire(R) Syndromic Trends dataset, which contains deidentified patient test data from BioFire Respiratory Panels collected at over 100 sites across the globe since 2015. Adenovirus Detected test results in BioFire Syndromic Trends were classified to produce predicted prevalence of each HAdV species within the United States from 2018 through 2021. These results show a marked change in both the predicted prevalence for HAdV and the species makeup with the onset of the COVID-19 pandemic. In particular, HAdV-B decreased from a pre-pandemic predicted prevalence of up to 40% to less than 5% in 2021, while HAdV-A and HAdV-F species both increased in predicted prevalence. Author SummaryIn the diagnosis of infectious disease there is often a trade-off between time to result, cost, and the amount of information gained about the pathogen. We develop a method that can provide increased genotypic information about a pathogen in a test result, with no additional time or cost. Our method is applied to the BioFire Respiratory(R) 2.1 Panel and the BioFire Respiratory(R) 2 Panel to speciate positive test results for adenovirus. The method shows robust performance in correctly identifying the adenovirus species from the test results. We estimate the prevalence of each adenovirus species in the United States by applying this method to adenovirus positive test results in the BioFire Syndromic Trends network. This information can be used in near real-time to augment the Centers for Disease Control and Prevention efforts to monitor adenovirus species circulation. Our mathematical framework can be generalized to other genotyping applications.

molecular biology↗

bioRxiv2018

A High-Quality De Novo Genome Assembly from a Single Mosquito using PacBio Sequencing

A high-quality reference genome is a fundamental resource for functional genetics, comparative genomics, and population genomics, and is increasingly important for conservation biology. PacBio Single Molecule, Real-Time (SMRT) sequencing generates long reads with uniform coverage and high consensus accuracy, making it a powerful technology for de novo genome assembly. Improvements in throughput and concomitant reductions in cost have made PacBio an attractive core technology for many large genome initiatives, however, relatively high DNA input requirements ([~]5 {micro}g for standard library protocol) have placed PacBio out of reach for many projects on small organisms that have lower DNA content, or on projects with limited input DNA for other reasons. Here we present a high-quality de novo genome assembly from a single Anopheles coluzzii mosquito. A modified SMRTbell library construction protocol without DNA shearing and size selection was used to generate a SMRTbell library from just 100 ng of starting genomic DNA. The sample was run on the Sequel System with chemistry 3.0 and software v6.0, generating, on average, 25 Gb of sequence per SMRT Cell with 20 hour movies, followed by diploid de novo genome assembly with FALCON-Unzip. The resulting curated assembly had high contiguity (contig N50 3.5 Mb) and completeness (more than 98% of conserved genes are present and full-length). In addition, this single-insect assembly now places 667 (>90%) of formerly unplaced genes into their appropriate chromosomal contexts in the AgamP4 PEST reference. We were also able to resolve maternal and paternal haplotypes for over 1/3 of the genome. By sequencing and assembling material from a single diploid individual, only two haplotypes are present, simplifying the assembly process compared to samples from multiple pooled individuals. The method presented here can be applied to samples with starting DNA amounts as low as 100 ng per 1 Gb genome size. This new low-input approach puts PacBio-based assemblies in reach for small highly heterozygous organisms that comprise much of the diversity of life.

genomics↗

bioRxiv2017

Resolving the Full Spectrum of Human Genome Variation using Linked-Reads

Large-scale population based analyses coupled with advances in technology have demonstrated that the human genome is more diverse than originally thought. To date, this diversity has largely been uncovered using short read whole genome sequencing. However, standard short-read approaches, used primarily due to accuracy, throughput and costs, fail to give a complete picture of a genome. They struggle to identify large, balanced structural events, cannot access repetitive regions of the genome and fail to resolve the human genome into its two haplotypes. Here we describe an approach that retains long range information while harnessing the advantages of short reads. Starting from only [~]1ng of DNA, we produce barcoded short read libraries. The use of novel informatic approaches allows for the barcoded short reads to be associated with the long molecules of origin producing a novel datatype known as Linked-Reads. This approach allows for simultaneous detection of small and large variants from a single Linked-Read library. We have previously demonstrated the utility of whole genome Linked-Reads (lrWGS) for performing diploid, de novo assembly of individual genomes (Weisenfeld et al. 2017). In this manuscript, we show the advantages of Linked-Reads over standard short read approaches for reference based analysis. We demonstrate the ability of Linked-Reads to reconstruct megabase scale haplotypes and to recover parts of the genome that are typically inaccessible to short reads, including phenotypically important genes such as STRC, SMN1 and SMN2. We demonstrate the ability of both lrWGS and Linked-Read Whole Exome Sequencing (lrWES) to identify complex structural variations, including balanced events, single exon deletions, and single exon duplications. The data presented here show that Linked-Reads provide a scalable approach for comprehensive genome analysis that is not possible using short reads alone.

genomics↗