Silica Corpora provides an AI-driven platform that designs therapeutic antibodies directly from amino‑acid sequences. Its modular system generates heavy and light chain candidates, predicts developability attributes, optimizes sequences to meet target product profiles, and maps antigen epitopes without requiring 3D structural data, enabling biopharma teams to accelerate candidate discovery and reduce experimental screening costs.
Funding
Funding not disclosed
Founders
Product
Problem
Developing therapeutic antibodies traditionally requires extensive experimental screening of large libraries, high costs, and long timelines, limiting the ability to rapidly generate candidates with optimal developability and target specificity.
Solution
Silica Corpora offers an AI-driven de novo antibody design platform that operates solely on amino‑acid sequences. Its modular system includes a Generator that creates heavy and light chain sequences, a Discriminator that predicts developability attributes such as stability, solubility, aggregation, and immunogenicity, an Optimizer that refines candidates to meet a program‑specific target product profile, and an Ep‑Mapper that predicts linear and conformational epitopes without needing 3D structures. By integrating project‑specific data from in‑vitro workflows, the platform delivers high‑accuracy predictions with minimal input data, accelerating candidate identification and reducing the number of experimental candidates required for pre‑clinical testing.
Target Audience
Primary customers are biopharmaceutical companies and research organizations developing therapeutic antibodies who need rapid, data‑driven candidate generation and optimization.
Features
- Generator module produces novel antibody heavy‑ and light‑chain sequences directly from target amino‑acid inputs
- Discriminator predicts key developability metrics (thermal stability, solubility, self‑aggregation, immunogenicity) with high accuracy
- Optimizer adapts generated sequences to meet defined target product profiles, outputting mutated candidates with improved predicted characteristics
- Ep‑Mapper predicts antigen epitopes (linear and conformational) independently of 3D structural data
- Protein Large Language Models (pLLMs) trained on millions of protein‑protein interactions and fine‑tuned on antibody‑antigen data for precise sequence generation
- Requires only amino‑acid sequence inputs; no additional equipment, expertise, or in‑vitro assays needed for the AI workflow
- Capable of leveraging small project‑specific datasets (≈100 candidates) to achieve accurate predictions, reducing data‑volume requirements