Researchers at the Indian Institute of Technology, Madras, have developed a ‘machine learning pipeline’, which they call ‘AI-expert’, for suggesting the best combination of amino acids to form peptides for a given purpose.

Amino acids are a group of 20 compounds of carbon, oxygen, hydrogen and nitrogen. Short chains of amino acids are ‘peptides’, while long chains are proteins.

Theoretically there are billions of peptides, because amino acids can, mathematically, combine in so many ways. Much like one can form countless words with the 26 letters of the alphabet.

Scientists have all along relied on experience and intuition to come up with newer peptides, but with artificial intelligence the selection of the right candidate to solve a particular problem can be more precise.

Prof Rohit Batra and his team at the Department of Metallurgical and Materials Engineering, Indian Institute of Technology, Madras, have developed the ‘AI-expert’ to pick the ‘right horses for the courses’ among the billions of peptides. The AI-expert uses an algorithm known as Monte Carlo tree search (MCTS) to make an informed decision on the peptide sequences.

Also read
Calotropis procera is commonly known as ‘apple of sodom’

Natural emulsifier from ‘apple of Sodom’

Bacteriophage therapy has great potential as an alternative to antimicrobials

Dealing with drug-resistant pathogens 

The 3D printing technology has helped accelerate research involving bio-inspired materials

More bio-inspired functional materials pouring out of labs 

To put it simply, the AI would come up with a peptide sequence. A computer would simulate its use, determine how good it is, and give it a score. Then the AI would come up with the next one and get a score. The scores will tell the AI which direction to take and, soon enough, it will come up with the right peptide.

For the research, Batra took up penta-peptides, or peptides of five amino acids, for which, mathematically, there are 3.2 million possibilities. The AI-expert evaluated 6,600 of these and compared the results given by a group of expert peptide designers. The AI-expert had a success rate of 66.7 per cent.

While the discovery of such peptides with special properties is by itself impactful, the work holds far greater potential, says Prof Rampi Ramprasad, a Georgia Research Alliance Eminent Scholar in Energy Sustainability, School of Materials Science and Engineering, Georgia Institute of Technology, Atlanta, Georgia. “It provides a blueprint for search strategies that could be adopted in many materials and chemistry application domains, so long as the search problem can be formulated in a suitable manner,” Ramprasad says in an article in IIT Madras Tech Talk.

Batra told Quantum that the AI-expert could be used wherever there is a need to choose from among billions of possibilities; he, however, chose to apply this with peptides — as opposed to, say, polymers — because peptides are easy to manufacture. One can come up with the right peptide for, say, light harvesting, catalysis, mechanical stability, or conductivity. You can, for example, come up with peptides that bind to certain rare earth materials, Batra said.

Can you use AI-expert to arrive at different proteins? Proteins can have hundreds of amino acid molecules, compared to peptides with 15-20 molecules. Not possible, says Batra. The computation required would overwhelm even the best computers of today. Proteins should wait for quantum computers.