Absci Investor Presentation Deck

DE NOVO DESIGN IN SILICO REQUIRES LOTS OF HIGH-QUALITY TRAINING DATA Highly validated ACE Assay generates high- quality and high-throughput data to train deep learning models 1. Strains expressing unique antibody sequence variants 2. Fix and permeabilize cells and add labeled probes ³00+ ¯¯TT Labeled scaffold-binding protein reports specifically on titer Labeled antigen reports on affinity 3. Screen and sort by flow cytometry 4. NGS Expression Binding M M M M Labeled probes Bachas, S., Rakocevic, G. et al., "Antibody optimization enabled by artificial intelligence predictions of binding affinity and naturalness," 2022 pre-print in bioRxiv. ABSCI CORPORATION 2023 ALL RIGHTS RESERVED 5. ACE Assay scores (binding classification) HCDR123 score 95.3% Metric Accuracy Precision Recall F1-score HCDR3 score 95.52% 95.39% 94.83% 0.9511 95.65% ● • ~800 positive controls 92.91% 0.92426 ~5K controls are spiked into libraries of 400k in size: • ~1000 negative controls ● Each included with multiple codon optimized variants absci. 13

View entire presentation