A research group headed by Huimin Zhao from the University of Illinois Urbana-Champaign have recently made some major breakthroughs developing mitochondrial targeting sequences (MTSs). They completed these feats with the help of generative artificial intelligence. The resulting research, now published in Nature Communications, reveals a groundbreaking approach to creating more complex MTSs. These MTSs are critical regulators of cellular homeostasis and possess significant therapeutic potential. Aashutosh Boob, a former doctoral student in Zhao’s group, serves as the first author of the publication titled “Design of diverse, functional mitochondrial targeting sequences across eukaryotic organisms using variational autoencoder.”
The mitochondrion is a key organelle. Cellularly, it plays a central role as an energy sensor in connective tissue growth factor and as a key metabolic regulator within the cell. In addition to its significance for basic research, targeting this organelle has received interest for therapies. For this powerful identification, the study team used an unsupervised deep learning structure known as Variational Autoencoder. They synthesized one million AI-derived MTSs, focusing both on their structural features and functional capacities.
Understanding Mitochondrial Targeting Sequences
Mitochondrial targeting sequences are highly conserved sequences that are critical for directing proteins to the mitochondria. These receptor diffusing sequences are on the order of 10-120 amino acids in length. On average, they are ~35 amino acids long. Important characteristics of MTSs are their positive charge, amphiphilic property, and propensity to adopt a α-helix conformation. Despite their significance, relatively few MTSs have been biochemically characterized.
Aashutosh Boob remarked that very few MTSs have been defined. Another point he made is that researchers are often guilty of annotating the same sequence over and over again. When attempting metabolic engineering, this redundancy is highly impactful. It frequently produces homologous recombination and genetic instability if the same sequences are implicated.
“One of the issues is that for different passenger proteins, there’s a different optimal targeting sequence. Secondly, if the same sequences are used often, particularly in metabolic engineering, it can actually lead to homologous recombination and then genetic instability. So ideally, there would be a library of diverse MTSs at hand to test and use.” – Aashutosh Boob
The ecological diversity of MTS is key to their widespread use in metabolic engineering and disease therapies. The study underscores the need for more diverse MTSs. This expansion will lead to better experimental results and it will greatly reduce the possibility of genetic instability.
Harnessing AI for Synthetic Biology
The researchers utilized cutting-edge AI technology to overcome the limitations imposed by conventional methods of researching MTSs. By leveraging Variational Autoencoder, the researchers were able to search complex output sequence spaces far beyond what is possible with traditional, design-heavy heuristic search methods.
Huimin Zhao is the Steven L. Miller Chair of Chemical and Biomolecular Engineering. We second him in pointing to the promises of AI to transform scientific research and discovery. AI is hot as fire right now. People are getting very excited about what things can be done with it, not the least of which is the scientific community. He continued, “This project is an excellent example of how generative AI can be a powerful tool for the synthetic biology and biotechnology fields.”
To rigorously determine if the AI-generated MTSs would properly target mitochondria, our research team experimentally tested 41 MTSs to evaluate their mitochondrial targeting abilities. This tangible use case serves to highlight the project’s role in advancing comprehension of mitochondrial biology and its therapeutic promise.
“Researchers want to study the biology of the mitochondria which can’t be done efficiently without using targeting sequences,” – Huimin Zhao
Aashutosh Boob’s Journey
Aashutosh Boob invested the better part of his Ph.D. tackling this project. Most importantly, it pressured him to expand his skillset away from classic lab research. This experience honed his skills in study design for scientific investigation. Perhaps most importantly, it provided him an opportunity to work with other talented, creative, driven minds in a responsive, energetic environment.
Assembling the targeting sequences in the lab ended up costing us a tremendous amount of time. We wanted to highlight their uses in metabolic engineering and therapeutics,” Boob added. His commitment to this new field is a bellwether for the increasing interest AI might be applied to helping biological sciences tackle complex problems.
The study represents a landmark advancement in our understanding of mitochondrial biology. It further lays bare the remarkable potential for generative AI to design novel functional biological tools. As our researchers work to uncover how best to tap into these exciting new pathways, the promise for new therapies and metabolic breakthroughs increases.