Adaptyv Bio EGFR Protein Design Competition

Sequences and additional screening data from Adaptyv Bio's EGFR Protein Design Competition

Round 1 submission

round1-egfr-inhibitors.fasta contains the original 10 sequences I submitted to the competition. They are described in this thread. One of them, gitter-yolo1, ranked 21 by iPAE and was selected for testing. It had low expression and did not bind EGFR in experimental testing.

The round1/llama-egfr-inhibitor-chat.* files are transcripts from the five Llama 3.1-405B designs.

Round 1 second submission

I made a post-round 1 submission to Adaptyv for additional experimental screening that included 11 sequences (round1-second-submission.fasta):

• The 9 from round1-egfr-inhibitors.fasta that were not selected by Adaptyv during the competition (all except gitter-yolo1, simply yolo1 here)
• Human EGF
• An additional sequence from @deepsatflow

Human EGF was suggested by @btnaughton and seconded by @gottapatchemall. It was of interest because many of the top-scoring designs by iPAE, including gitter-yolo1, were based on EGF but did not bind EGFR. The specific sequence proposed was a 53 amino acid sequence from UniProt. This was the same sequence I obtained from PDB 8HGS for my submission and confirmed in Bai 2023.

deepsatflow-design:7 n:0|mpnn:1.320|plddt:0.922|i_ptm:0.893|i_pae:4.337|rmsd:1.423 was created using methods in a separate GitHub repository. Briefly, it is a ColabFold workflow that involves RFdiffusion, ProteinMPNN, and AlphaFold-Multimer.

None of the designed round 1 second submissions bound EGFR. Only the positive control, human EGF, bound the target. See round1-second-submission-data.csv for the data table and the round1-second-submission-raw-data subdirectory for the raw kinetic data.

The Adaptyv team noted that they ran this experiment at a higher starting concentration (1000 nM) compared to the competition. Also, some of the sequences (e.g. gitter-yolo3 and gitter-yolo5) exhibited aggregation.

Round 2

The round2 subdirectory contains the complete strategy, code, and submissions for round 2 of the competition. gitter.yolo23, gitter.yolo46, and gitter.yolo48 were selected for experimental testing! All three expressed, and gitter.yolo48 was a weak binder. I shared thoughts on the round 2 results.

Additional information

• Adaptyv Bio round 1 results
• Adaptyv Bio round 1 GitHub dataset
• Adaptyv Bio round 1 competition lessons
• The BioLM team submitted additional sequences after round 1 and shared their analysis
• suzuki-2001's Adaptyv competition repo
• Adaptyv Bio round 2 submission analysis
• Adaptyv Bio round 2 results
• Adaptyv Bio round 2 GitHub dataset