AlphaFold 3 vs BANDOCK+: Benchmarking Drug Discovery Scores

bioRxiv

Overview

AlphaFold 3 (AF3) demonstrates potential for computational drug discovery, achieving slightly higher aggregate disease ranking scores than BANDOCK+ in a 1,103 protein comparison. However, BANDOCK+ provides significantly broader protein coverage and lower computational demand, utilizing a 20,158 protein reference library. This highlights a critical tradeoff between predictive performance and scalability in multiscale drug discovery.

In Depth

Key Findings

AlphaFold 3 (AF3) confidence scores show utility in predicting protein ligand complexes, with the AF3 ipTM metric achieving slightly higher observed aggregate disease ranking scores compared to BANDOCK+ in a matched comparison of 1,103 proteins. While AF3 shows strength in predictive performance, the results vary across individual indications.

Technical Details

The study benchmarked AF3 against BANDOCK+, an experimentally informed approach, using human protein libraries and approved drugs. A primary differentiator between the two methods is scalability and computational efficiency:

• AlphaFold 3: Demonstrated higher ranking scores in the 1,103 protein set but requires more computational resources.
• BANDOCK+: Enabled substantially broader protein coverage, utilizing a reference library of 20,158 proteins while remaining computationally less demanding.

Background & Context

While AF3 is capable of predicting protein ligand complexes, the specific utility of its confidence scores for practical drug discovery applications had remained unclear. This benchmarking effort sought to clarify the performance of AF3 relative to existing experimentally informed methods like BANDOCK+ to determine where each tool fits within a drug discovery pipeline.

Significance & Outlook

The findings establish a clear tradeoff between the high-precision predictive capabilities of AF3 and the scalability of BANDOCK+. For researchers and engineers, this suggests that while AF3 confidence scores are valuable for targeted predictions, BANDOCK+ remains more viable for large-scale protein coverage due to its efficiency. The results provide a framework for choosing between predictive performance and computational scalability depending on the scope of the drug discovery project.

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.

• Your email and selected fields are used only to deliver the newsletter.
• We never share your information with third parties.
• You can unsubscribe anytime via the link in each email.

See our Privacy Policy for details.

Agree & Continue ▶

Takes about a minute · Unsubscribe anytime