Michael Akinyemi
MIT Department: Chemical Engineering
Faculty Mentor: Prof. Bandon DeKosky
Research Supervisor: Shelbe Johnson
Undergraduate Institution: University of Central Florida
Hometown: Tallahassee, Florida
Website: Intern’s Website, LinkedIn
Biography
Michael Akinyemi is a rising senior at the University of Central Florida pursuing a B.S. in Biotechnology. His personal experiences with the shortcomings of modern medicine have instilled the belief that we need researchers with a truly interdisciplinary understanding of both computer science and molecular biology. His research shows that he strives to integrate both.
Under the DeKosky Lab at MIT, he improved their immune receptor sequencing platform by working to generate a cell line optimized for higher-efficiency library uptake, decreased the runtime of supercomputing analysis pipelines by >70%, and designed probabilistic models to detect next-generation sequencing errors. At his home institution, he independently developed scRNA-Seq analysis tools to study autoimmune disorders and investigated approaches to integrating multimodal datasets under the Nguyen Lab. You’ll often see him biking, hiking, and playing/developing video games. He loves to sit down and play with his two adorable guinea pigs to de-stress.
Abstract
Probabilistic Approaches to Correct Antibody Library Sequence Errors
Michael J. Akinyemi1,2, Yun-Ti Chen2, Shelbe Johnson2,3 and Brandon J. DeKosky2,3
1Department of Biomedical Sciences, University of Central Florida
2The Ragon Institute of MGH, MIT and Harvard
3Department of Chemical Engineering, Massachusetts Institute of Technology
The development of new vaccines, cancer therapies, and drugs often relies on the generation and screening of antibody libraries. By artificially introducing small mutations into these libraries, we can create new sets of antibody variants with diverse properties. These libraries are then screened to identify antibodies with desired characteristics, such as high binding affinity for potential drug targets. However, techniques like next-generation sequencing (NGS) often have significant error rates, leading to false-positive mutations in antibody sequence data. It is crucial that we prioritize accuracy when sequencing these libraries, as single nucleotide mutations can drastically alter antibody properties. While many error-correction tools already exist, they primarily target large genome sequencing and are not suited for the immense diversity of antibody display libraries. Here we present a new approach using probabilistic models to differentiate true variants from erroneous mutations in antibody gene libraries. We show that probability-based models can often efficiently detect mutations in sequence data based on prior knowledge of known error patterns. While this tool has been developed for antibody engineering applications, it also has potential utility for a wide range of library screening studies across molecular biotechnology.
