Mukund Sudarshan

Splicing of pre-messenger RNA (pre-mRNA) is an essential process necessary for proper gene expression in eukaryotic cells, and occurs in the vast majority of human genes. During this process, the pre-mRNA is transformed into a mature messenger RNA (mRNA). This involves the recognition of intron/exon boundaries, the subsequent excision, or splicing, of the introns, followed by joining of the adjacent exons. The splicing process is governed by an array of sequence features known as the splicing code. This project seeks to significantly advance the understanding of this splicing code using machine learning models and apply this knowledge towards the development of splice-switching antisense oligonucleotides (ASOs) in particular. ASOs, such as the FDA-approved drug Nusinersen, are short chemically-modified nucleic acids that modulate the splicing process. Knowledge of this splicing process and being able to predict the exon sequence from any pre-mRNA sequence can lead to significant improvements in both the diagnosis and treatment of genetic disorders. There is a strong unmet clinical need for such knowledge in order to create efficient drugs for antisense therapy, as currently, the design of ASOs is often determined by a search through a very large space of sequences. Using state-of-the-art machine learning techniques, this project will model the splicing code in silico, and then use these models to rationally design ASOs. Preliminary work in designing machine learning models based on data from massively parallel reporter assays (MPRAs) is very encouraging and has led to high prediction accuracy, and more importantly, several mechanistic insights into the splicing code. Further insights from these machine learning models will assemble a coherent picture of the splicing code and how it interacts with ASOs.