Off-target hit prediction for drug candidates, confirmed in a partner lab
TL;DR Sidebind helps small-molecule drug teams spot unwanted protein interactions early. It predicts which off-target proteins each candidate is likely to hit, ranks the safety risks and arranges lab assays for the top ones through a partner contract research lab. Indian biotechs and pharma R&D groups pay per compound series.
Most drug candidates fail in the clinic, and safety is a big reason why. A 2022 review of clinical data from 2010 to 2017 put 30% of failures down to unmanageable toxicity and 40–50% to lack of efficacy[1]. Some of that toxicity comes from a molecule binding proteins it was never meant to touch.
Large pharma companies screen candidates against standard off-target panels early, a practice set out in a widely cited 2012 review[2]. Smaller Indian biotechs and generic companies moving into new molecules often test late, or test a narrow panel, because every assay costs money and time and that adds up fast. A problem found at the candidate stage wastes months of chemistry.
What it does, in regulatory terms: research decision support for preclinical chemistry only. Sidebind's predictions are not a regulatory safety study and do not replace the nonclinical data that the New Drugs and Clinical Trials Rules 2019 require before human trials[3].
A medicinal chemist uploads a compound series. Sidebind's models, trained on public bioactivity data such as ChEMBL[4] plus the results of its own partner assays, estimate the chance that each compound binds known safety-relevant proteins. The report ranks the risks and recommends a small, targeted assay set. A partner contract lab runs those assays, and the results flow back to retrain the models.
The seed proposed full polypharmacology prediction across all proteins. Today's models aren't accurate enough to stand alone, so I reframed it as triage - predict, then confirm the few hits that matter in a lab. Sidebind reports its hit rate against lab results openly.
Your buyer is a head of discovery chemistry or a safety pharmacology lead at an Indian biotech, a mid-sized pharma company with an innovation pipeline, or a contract research organisation that wants to offer the service. Pricing per compound series plus assay pass-through suits small teams (our estimate). Big-company assay panels are sold on quote, so there's no public benchmark price.
Indian AI-led discovery companies are now doing deals. One Bengaluru AI discovery startup has licensed a repurposed candidate to a partner for Duchenne muscular dystrophy[5]. Public bioactivity data keeps growing, and methods for compound-protein interactions have improved enough that triage, rather than replacing the lab, is a realistic goal.
There's no public figure for Indian spending on off-target safety profiling or in-silico safety prediction. The signal is indirect - the share of clinical failures linked to toxicity[1], and Indian discovery firms licensing assets and selling AI-led R&D services[5][6]. Treat the global drug-discovery outsourcing market as context only.
One Indian firm sells computational chemistry, virtual screening and custom AI models to biopharma R&D teams[6]. Another runs its own AI discovery and repurposing pipeline[5]. Global contract labs sell standard off-target panels. None of the Indian firms whose sites I read offers off-target prediction tied to lab confirmation as a packaged product, and that combination is the gap.
| Difficulty | Hard (56/100) |
| Starting capital | ₹2.5–4.5 Cr |
| Time to launch | 12–18 months |
The science is the hard part. Public data is noisy, and the models have to be honest about where they're unsure. You also need a trusted lab partner and the patience to win over chemists who distrust black-box scores. Client compounds are crown-jewel data, so security has to be strong from day one.
The software itself needs no licence. The rules matter for how clients use your output and for your lab partner.
Building this outside India? The categories above usually have a local equivalent, so check with your own regulator.
AI here is research triage, not a safety study. Never let a prediction replace the nonclinical data a regulator requires.
Want to work out the AI layer for Sidebind? Talk to Yash →
This suits a founding team with a computational chemist and a safety pharmacologist who've shipped models used in real programmes. If you're a generalist AI founder then honestly, it's the wrong idea. First prove the core number on blinded historical compounds from one partner - how often the top-ranked predictions match assay results.
Plan for roughly ₹2.5–4.5 Cr, mostly on a computational chemistry team, data and partner-lab assays.
Not for the software. Your outputs can't replace the nonclinical studies required under the New Drugs and Clinical Trials Rules 2019, and your lab partner needs proper quality systems.
Yes, discovery teams everywhere screen for off-target risk. Check your local drug regulator's rules on nonclinical data and the licence terms of every dataset you use.
The free Sidebind validation kit has the line-by-line capital breakdown, the effort scoring, the data-licensing and regulatory checklist, a 14-day validation sprint with pass/fail targets, a 90-day launch plan, 10 interview questions for discovery teams, and 5 assumptions to test. Download the kit →
Thinking about building Sidebind or something close to it? The make-or-break work is a blinded benchmark that discovery chemists actually believe, and I can help you plan it and the raise that follows. If you need a hand, you know where to find me. Talk to Yash →
For illustration only. Names are fictional and ideas hypothetical; sourced facts are real but may be out of date. Nothing here is legal, financial or professional advice, and Yash Khatri is not personally liable for any use of it. Read the full disclaimer.