From 40,000 Molecules to One, in Five Months

What Actually Happens When AI "Screens 40,000 Molecules"

Here's what's actually happening underneath that number: a protein's PDZ domain is a shallow, broad surface that interacts with many other proteins loosely, rather than having one deep pocket a drug molecule can lock into precisely - which is exactly why GIPC1 sat in the "undruggable" category for years.

Generative AI's job here wasn't searching a library of existing drugs; it designed entirely new candidate molecular structures from scratch, shaped specifically to fit that awkward, shallow binding surface. Predictive AI then did the second job - forecasting which of those generated candidates would likely be too toxic, poorly absorbed or otherwise unviable, before anyone spent months synthesising them in a lab.

That two-step division of labour - generate, then predict-and-filter - is the actual mechanism that turned 40,000 possibilities into two molecules worth physically testing.

Why the Compression Matters More Than the Molecule

Traditional drug discovery for a genuinely novel target typically takes years just to reach the "molecule worth testing" stage, because each round of synthesis-and-test is slow and expensive and researchers can only chase a handful of chemical hypotheses at a time.

AI collapses that constraint by testing chemical hypotheses computationally - via molecular modelling and quantum chemistry, simulating how candidate molecules would actually bind to GIPC1's PDZ domain - before committing lab resources to synthesis.

Sravathi AI's five-month discovery phase for GIPCi, versus the years such work traditionally requires, is the number that actually captures the shift: this isn't a story about one drug candidate, it's a demonstration that "undruggable" targets across cancer biology are now economically worth attempting, because the cost of exploring the chemical search space computationally is a fraction of exploring it physically.

Where India Actually Stands

This case places India in a specific, meaningful position: not as a downstream consumer of AI drug-discovery tools built elsewhere, but as a source of the computational platform itself, developed by a Bengaluru startup and licensed into a partnership with one of the world's leading medical research institutions under a shared intellectual-property arrangement.

That is a different kind of participation in global biotech than clinical trial hosting or generic manufacturing, India's more traditional roles - this is upstream, in the actual discovery layer, which is where the highest-value intellectual property in pharmaceuticals typically sits.

For an aspirant, the exam-relevant insight isn't "AI helped find a cancer drug"

• it's the specific two-stage mechanism (generative design, then predictive filtering) that makes previously undruggable protein targets tractable and the fact that an Indian company built and owns a meaningful share of that capability rather than merely applying someone else's.