A multi-omic machine-learning framework for stress-aware therapeutic protein biomanufacturing - efficient protein production underlies applications ranging from biologics manufacturing to…

The National Institute of Biomedical Imaging and Bioengineering awarded Massachusetts Institute of Technology $100,228 on January 30, 2027, under the Discovery and Applied Research for Technological Innovations to Improve Human Health program (CFDA 93.286) to develop a multi-omic machine-learning framework for stress-aware therapeutic protein biomanufacturing.

The project will generate high-resolution multi-omics datasets including RNA-seq, quantitative proteomics, tRNA abundance, and tRNA modification profiles across bioreactor-relevant stresses in Komagataella phaffii, a major industrial protein production host. MIT will integrate these measurements into structure-aware machine-learning models that incorporate genomic codon distributions, mRNA structure, and position-specific features within encoded protein three-dimensional architecture. The team will experimentally test model predictions by synthesizing and expressing codon-designed variants of FGF2, a commercially and biologically important growth factor, using MIT's automated high-throughput screening capabilities. Iterative model refinement through active learning will yield a generalizable platform for rational, physiology-aware codon design to optimize translation efficiency under specific growth and stress conditions.

Performance occurs in Cambridge, Massachusetts, with an ultimate completion date of August 31, 2028.

A multi-omic machine-learning framework for stress-aware therapeutic protein biomanufacturing - efficient protein production underlies applications ranging from biologics manufacturing to sustainable food systems, yet current codon optimization tools remain limited by outdated assumptions such as static trna pools and uniform codon preferences. Recent discoveries from the dedon lab demonstrate that cells dynamically reprogram trna abundance and chemical modification states in response to stress, producing coordinated shifts in synonymous codon usage that selectively enhance translation of stress-response gene families. These mechanisms are particularly relevant in komagataella phaffii, a major industrial host that experiences substantial translational stress in bioreactor environments. However, no existing framework integrates stress-responsive trna dynamics, RNA structure, and protein structural context to guide codon design. This fellowship project aims to develop a multilayered, machine-learning platform that predicts synonymous codon usage patterns optimized for high translation efficiency under specific growth and stress conditions. Aim 1 will generate high-resolution multi-omics datasets- including RNA-SEQ, quantitative proteomics, trna abundance, and trna modification profiles-across bioreactor-relevant stresses in k. Phaffii. Aim 2 will integrate these measurements into structure-aware ML models that incorporate genomic codon distributions, mrna structure, and position-specific features within the encoded protein's 3-d architecture. Aim 3 will experimentally test model predictions by synthesizing and expressing codon-designed variants of FGF2, a commercially and biologically important growth factor, using MIT's automated high-throughput screening capabilities. Iterative model refinement through active learning will yield a generalizable platform for rational, physiology-aware codon design. The training plan combines rigorous experimental work in multi-omics and RNA chemistry with computational training in machine learning, protein structure analysis, and predictive modeling. Mentorship, coursework, and the interdisciplinary environment of the MIT biological engineering department will prepare the applicant for an independent research career focused on translational control, RNA biology, and data-driven approaches to biomanufacturing. The resulting framework will advance both fundamental understanding of codon-biased translation and practical strategies for improving protein production in industrial and biomedical settings.