Cracking the Protein AI “Black Box”: Neel Mukavilli’s Research on Interpretable Protein Design
Neel Mukavilli, a senior Bioengineering major and Chemistry minor, is using high-performance computing to solve one of the most pressing challenges in modern biotechnology: making Artificial Intelligence understandable. Working in Dr. Michelle McCully’s lab, Neel is not just using AI to design new proteins; he is "probing" the very internal logic of these models to turn accidental discoveries into intentional design.
The Challenge: Turning Accidental Stability into Intentional Design
AI models are incredibly proficient at designing protein sequences for therapeutics, such as insulin. However, these programs often function as "black boxes"—researchers receive a design but don't understand the reasoning behind the AI's choices.
"We found that the AIs that design proteins... all end up making proteins very stable even at higher temperatures," Neel explains. While high-temperature stability is often a desirable trait, currently, the AI achieves it "on accident". Neel’s research aims to solve the problem of increasing interpretability in AI programs so that these traits can be engineered on purpose.
How WAVE HPC Powers the Research
Neel’s day-to-day workflow relies heavily on the WAVE HPC cluster, which he uses for two primary computational tasks:
- AI Training: Neel utilizes Python and PyTorch to train Sparse Autoencoders (SAE). These are smaller AI models nested inside a larger program called Protein MPNN (a message passing neural network) to "probe" its internal decision-making layers.
- Molecular Dynamics (MD) Simulations: Using NAMD software and WAVE’s specialized bio nodes, Neel runs simulations that map the 3D coordinates of every atom in a protein. These simulations track how a protein reacts to forces over 100,000 frames, a task that requires significant processing power.
Image: Protein dynamics simulation visualizing atomic movements over time, run on the WAVE HPC cluster.
"WAVE also makes it a lot easier. I don't need to go out and buy my own computer," Neel says, noting that the cluster allows him to run longer simulations and get better data than would be possible on standard hardware.
Bridging the Gap Between Computer and Wet Lab
For Neel, the HPC cluster is an essential tool for speeding up biological research. While computational methods will never fully replace "bench chemistry," they allow researchers to narrow down a list of candidates much faster than traditional wet lab testing.
Simulating a protein’s reaction to high temperatures on a computer is significantly more efficient and cost-effective than physically acquiring the protein and testing it with reagents like urea. By using the HPC to identify the best candidates first, researchers can drastically reduce testing time and costs for new medical treatments.
A Foundation for the Future
Neel’s work at SCU has already garnered significant recognition; he was the recipient of the ALZA Science Scholars Research Award and the SCU WAVE HPC grant. What began as an individual project has now expanded into a senior design project where Neel leads a team of three other students.
As he prepares to graduate and pursue a PhD in Biochemistry or Bioengineering, Neel credits his time on the WAVE cluster with giving him a competitive edge. "I can’t be thankful enough for the opportunity I’ve had to do this research at SCU because it’s helped me develop my work and connect with professors at other universities."
Advice for Fellow Students
For students hesitant about the technical hurdles of high-performance computing, Neel’s advice is simple: try it out. With the addition of Jupyter Lab notebooks on WAVE, Neel emphasizes that the resource is incredibly accessible. "If you have ever had any experience coding in Python, [it is] very easy to get started".
Based on an interview conducted by Ella Griffin, WAVE Student Marketing Assistant, February 11, 2026.