Biology subjects

Lumetta, S. S.

Publications and source records attributed to Lumetta, S. S..

2 recordsLinked to original sources

bioRxiv2023

PandoGen: Generating complete instances of future SARS-CoV2 sequences using Deep Learning

One of the challenges in a viral pandemic is the emergence of novel variants with different phenotypical characteristics. An ability to forecast future viral individuals at the sequence level enables advance preparation by characterizing the sequences and closing vulnerabilities in current preventative and therapeutic methods. In this article, we explore, in the context of a viral pandemic, the problem of generating complete instances of undiscovered viral protein sequences, which have a high likelihood of being discovered in the future using protein language models. Current approaches to training these models fit model parameters to a known sequence set, which does not suit pandemic forecasting as future sequences differ from known sequences in some respects. To address this, we develop a novel method, called PandoGen, to train protein language models towards the pandemic protein forecasting task. PandoGen combines techniques such as synthetic data generation, conditional sequence generation, and reward-based learning, enabling the model to forecast future sequences, with a high propensity to spread. Applying our method to modeling the SARS-CoV-2 Spike protein sequence, we find empirically that our model forecasts twice as many novel sequences with five times the case counts compared to a model that is thirty times larger. Our method forecasts unseen lineages months in advance, whereas models 4x and 30x larger forecast almost no new lineages. When trained on data available up to a month before the onset of important Variants of Concern, our method consistently forecasts sequences belonging to those variants within tight sequence budgets. PandoGen is available at: https://github.com/UIUC-ChenLab/PandoGen

bioinformatics↗

bioRxiv2020

HELLO: A hybrid variant calling approach

Next Generation Sequencing (NGS) technologies that cost-effectively characterize genomic regions and identify sequence variations using short reads are the current standard for genome sequencing. However, calling small indels in low-complexity regions of the genome using NGS is challenging. Recent advances in Third Generation Sequencing (TGS) provide long reads, which call large-structural variants accurately. However, these reads have context-dependent indel errors in low-complexity regions, resulting in lower accuracy of small indel calls compared to NGS reads. When both small and large-structural variants need to be called, both NGS and TGS reads may be available. Integration of the two data types with unique error profiles could improve robustness of small variant calling in challenging cases. However, there isnt currently such a method integrating both types of data. We present a novel method that integrates NGS and TGS reads to call small variants. We leverage the Mixture of Experts paradigm which uses an ensemble of Deep Neural Networks (DNN), each processing a different data type to make predictions. We present improvements in our DNN design compared to previous work such as sequence processing using one-dimensional convolutions instead of image processing using two-dimensional convolutions and an algorithm to efficiently process sites with many variant candidates, which help us reduce computations. Using our method to integrate Illumina and PacBio reads, we find a reduction in the number of erroneous small variant calls of up to ~30%, compared to the state-of-the-art using only Illumina data. We also find improvements in calling small indels in low-complexity regions.

bioinformatics↗