The development of antibody-based drugs begins with discovering an antibody capable of binding to a specific site, known as an epitope, on a disease-related protein. Finding such a suitable antibody can be challenging.
ImmunitoAI, a Bangalore-based company, applies artificial intelligence to solve this problem. Its platform, AbInitio, is designed to engineer antibodies for a chosen target and epitope, after which the startup conducts testing in its own laboratory.
The startup was founded in 2020 by Dr. Aridni Shah and Trish Chatterji through the Entrepreneur First talent accelerator, where they met in August of that year.
Shah, who serves as CEO, holds a PhD in Biological Sciences from NCBS-TIFR and a Bachelor's degree in Biomedical Sciences from the University of Delhi. Chatterji, the CTO, has a Master's degree in Computer Science from the University of Delhi and previously worked as a senior software development engineer in the R&D department at GreyOrange.
The initial idea stemmed from Shah's doctoral research on honeybees. Since honeybees are not a model organism, obtaining antibodies against their proteins was difficult. Early discussions with industry experts indicated that drug manufacturers face a similar challenge when searching for suitable starting antibodies, notes Chatterji.
Chatterji spent about three months developing the idea and up to three weeks creating the first prototype. This version ranked potential antibodies, which is a simpler task than the antibody design work that AbInitio performs today.
ImmunitoAI's biology team first defines the disease area, the target protein, and the epitope to which the antibody must bind. It also determines whether the antibody should block or activate the target. Then, AbInitio designs antibodies and antibody-like structures, including nanobodies—small, single-domain antibodies—for the selected epitope.
The platform is trained on the three-dimensional structures of binding protein pairs from the Protein Data Bank, a public repository. A mathematical transformation developed by the startup converts the three-dimensional atomic coordinates into two graphs, where atoms are represented as points and bonds show their interaction sites.
The model studies how the shapes of two proteins fit together, with grooves on one corresponding to bumps on the other. It also examines which amino acids, the building blocks of proteins, bind to each other. During the design phase, the platform checks drug-like properties to reduce the need for subsequent refinement.