Absci Investor Presentation Deck

CASE STUDY: DE NOVO DESIGN IN SILICO Al designs of all HCDRS achieve high binding rates and outperform biological baselines Zero-shot de novo generated Human HER2 Rat HER2 HER3 Matched input antigen Mis-matched input antigen Mis-matched input antigen Mis-matched input antigen ABSCI CORPORATION 2023 ALL RIGHTS RESERVED VEGF Biological baseline OAS OAS-J SAbDab HER2 Binding Rate (%) measured via ACE assay HCDR3 HCDR123 Random baseline Permuted sequences 10.6 2.8 2.9 2.5 2.68 5.25 3.16 0.33 1.8 0.5 0.2 0.0 0.16 0.32 0.06 N/A Al designs are specific Inputting a mis-matched undesired antigen (e.g., Rat HER2, HER3, VEGF) into the model results in significant performance decrease towards desired antigen ● ● Al models outperform biological baselines De novo designed HCDR3s achieve a 4-fold improvement over random OAS baseline ● Indicates the model's use of antigen information for sequence designs ● De novo designed HCDR123s achieve an 11-fold improvement over random OAS baseline absci. 15

View entire presentation