A Partnership with Purpose

Working closely with Ellison Medical Institute (EMI) and under the guidance of EMI Director and cancer researcher Dr. David Agus, our joint team is tackling significant milestones in cancer research and novel therapeutic development. Together, we've already developed advances in AI-driven drug discovery, most notably an optimized implementation of AlphaFold2—the AI system for predicting protein structures—that dramatically accelerates the drug discovery process. Through this ongoing collaboration, a successful workflow was built and implemented in a matter of weeks 

The Challenge of Protein Structure Prediction

Understanding protein structures is a bit like having a blueprint for life itself. These structures are crucial for developing new medicines, but predicting them has historically been like solving an enormously complex 3D puzzle that could take hours or even days to complete. AlphaFold2 has revolutionized this field by making accurate predictions possible, but there was still a significant barrier—computational time. 

Salt’s Breakthrough with AlphaFold2 Optimization

When we started working with the EMI research team, we saw brilliant scientists bottlenecked by technology throughput limitations. By using Salt’s platform, we have been able to generate processing times 22x faster than previous benchmarks running on the same GPU hardware while maintaining full prediction accuracy. 

But speed isn't the whole story.

Transparent Architecture by Design

What makes our approach to AI-driven research unique is its transparency. In drug discovery, you can't afford to work with black box systems where you are unable to see how decisions are being made. Salt is built to provide complete visibility into every step of the discovery pipeline. Researchers can trace exactly how predictions are made, validate each step, and fine-tune the process to meet their specific needs.

The technical highlights behind this achievement include:

• A split-compute architecture that optimally balances CPU and GPU processes

• A sophisticated caching system for our 3TB dataset

• GPU-accelerated search capabilities that transform hours-long searches into seconds

• The ability to swap, mix, and combine AI models into workflows

• Simplified algorithmic search methods that speed up iteration cycles

Making Continuous Advancements from Discovery to Market

This step is just the beginning in a longer, impactful journey for the biotechnology industry and life sciences field. Now, we are improving diffusion model performance—a step notorious for its computational intensity. This reflects our broader company vision of creating a life sciences-dedicated AI workflow platform where researchers are limitless in their model use, including solutions that span the drug development lifecycle. While the technical challenges are significant, the potential impact on patients' lives makes every obstacle worth overcoming.

To learn more about our work in the life sciences or explore how Salt could support your research team, please visit salt.ai.