Rafael García Meseguer from ProtoQSAR, July 2026
How we flag unsafe ingredients before they reach the bench, so formulators can design safer lubricants faster
A lubricant is a mixture of ingredients, each a chemical that comes into contact with people and the environment. Designing a greener formulation starts with a simple question: is this molecule safe, and will it break down once it leaves the machine? Testing every candidate in the lab is slow and costly. By the time the results arrive, a formulator may have spent months on a compound that was never going to pass.
Answering that question early changes the economics of the whole project. If we can tell which ingredients are likely to be toxic or persist in the environment before any test runs, a formulator can drop the weak candidates on day one and spend the lab budget where it counts. The result is safer lubricants reaching the market sooner, with fewer months sunk into chemistry that was never going to work.
That’s the problem we address in SITOLUB. We use QSAR (Quantitative Structure-Activity Relationship) models to estimate a molecule’s hazard from its chemical structure before anyone synthesises or tests it. A molecule’s structure governs its behaviour, so with enough measured data on similar structures we can build a statistical model that predicts how a new one will behave. We apply it across the endpoints that matter for ecofluids: human health hazards, aquatic toxicity, environmental fate, and the physicochemical properties behind them.
In practice, a formulator sends us a set of candidate structures. We run them through the models and return a hazard profile for each, flagging, for example, those likely to harm humans or aquatic life, or to resist biodegradation. The models act as a filter ahead of the experiments, showing what’s worth measuring. They catch obvious problems early, when changing course is cheap, and leave the final word to experiment.
We always state one caveat up front: a QSAR model is reliable only within its applicability domain, the chemical space covered by the data it was trained on. If a molecule bears no resemblance to that training set, the prediction is weak. So every result we provide includes a domain check and a confidence flag. When a molecule falls outside the domain, we say so and disregard the prediction.
In SITOLUB, we build and validate our models on the chemistry that actually occurs in lubricants. The training sets hold the base oils, additives and thickeners common in real formulations, so the molecules a formulator is likely to ask about fall within the domain rather than at its edge. Where data is thin for a given family, we flag it and fill the gap with new measurements from our partners.
The predictions don’t stand alone. In SITOLUB, they integrate the project’s broader platform with other computational tools such as molecular dynamics, tribological simulation, and life cycle assessment, and support the AI-assisted design of safer alternatives.
The chemist stays in the loop. We give them a faster way to ask, “Is this worth pursuing?” with a documented reason for the answer. For a project designing lubricants that are safe and sustainable from the start, that head start counts.