Accelerate Drug Discovery with SPRINT on Vecura

What is SPRINT (panspecies-dti)?

SPRINT (Structure-aware Protein-ligand INTeraction) is a cutting-edge deep-learning framework designed to revolutionize drug-target interaction (DTI) prediction and large-scale virtual screening. By jointly embedding protein sequences and ligand fingerprints into a shared latent space, it enables researchers to predict interaction strengths using simple cosine similarity. This approach effectively replaces computationally expensive 3D docking simulations with a single dot product, allowing for the screening of millions or billions of candidate small molecules in mere minutes on standard GPU hardware.

What can users do with SPRINT on Vecura?

With SPRINT on Vecura, users can:

• Perform Large-Scale Virtual Screening: Rapidly rank massive libraries of small molecules against a specific protein target.
• Generate Protein Embeddings: Encode protein sequences into a latent representation, suitable for downstream tasks like vector-database search or similarity analysis.
• Encode Ligands: Convert small molecule SMILES strings into structured latent space vectors for high-dimensional analysis.
• Flexibly Switch Featurization: Choose between structure-aware screening (using SaProt with Foldseek) for maximum accuracy or sequence-only screening (using ProtBert) when structural data is unavailable.

What the output means

The output provides a ranked list of candidate molecules based on their predicted interaction strength in the SPRINT latent space, alongside high-dimensional embedding vectors.

It is crucial to understand that the interaction score is a relative ranking signal and not a calibrated binding affinity (such as Kd, Ki, or IC50). This output should be used to intelligently prioritise candidates for further computational validation or experimental assays; it does not replace the necessity for laboratory verification.

Why this matters

The traditional bottleneck in computational drug discovery is the sheer cost and time associated with docking large libraries of compounds against high-resolution protein structures. By enabling "virtual screening at the speed of a single dot product," SPRINT dramatically lowers the barrier to entry for identifying lead compounds.

This capability allows researchers to explore broader chemical spaces more efficiently, facilitating faster initial screening cycles. By integrating this technology into a streamlined workflow, teams can focus their experimental resources on the most promising candidates, significantly accelerating the early stages of the drug discovery pipeline.

• Developed by: Andrew McNutt, Abhinav Adduri, Caleb Ellington, and Monica Dayao.
• Source: Official GitHub Repository
• Reference: Original Paper (arXiv 2411.15418)