From Poison to Algorithm: What Insilico Has to Teach Brazil About Radical Innovation

Last week, Insilico Medicine announced the start of Phase III trials for rentosertib, an oral TNIK inhibitor for idiopathic pulmonary fibrosis (IPF). At first glance, the news seems like just another biotechnology milestone among many. But it's unlike anything we've seen so far, and it's worth understanding why before we return to the discussion about the Brazilian program for radical innovation in healthcare.

Rentosertib is the first molecule in history to reach a pivotal Phase III study having been developed entirely through artificial intelligence at every stage of the process. The therapeutic target, the TNIK protein, was identified by PandaOmics , Insilico's own computational biology platform. The molecule's chemical structure was designed by generative AI using Chemistry42 , another proprietary engine from the company. And the design of the clinical trial itself, including outcome prediction, was guided by inClinico , a third layer of artificial intelligence dedicated to anticipating clinical results. It's not a molecule discovered by traditional methods and then optimized with the help of AI. It's a molecule that, from the first to the last step, would not exist without artificial intelligence.

The science behind the Insilico milestone

Idiopathic pulmonary fibrosis is a chronic, progressive disease with no treatment yet capable of reversing its course, in which lung tissue scars and stiffens irreversibly. Median survival after diagnosis is usually between two and four years , highlighting the medical urgency behind this program. In the Phase IIa results, published in Nature Medicine and presented at the American Thoracic Society congress in 2025, the 60 mg once-daily arm showed an average improvement of 98.4 mL in forced vital capacity in just twelve weeks, with a manageable safety profile. The new Phase III trial, conducted by leading researchers from Peking Union Medical College and Shanghai Pulmonary Hospital , will follow 320 patients for a full year in a randomized, double-blind, placebo-controlled design.

It is this last detail that deserves close attention from those who follow the sector. Historically, between 35% and 50% of Phase III studies fail , even when previous phases showed strong signs. The question that Insilico and the pharmaceutical market will answer in the coming months is no longer whether artificial intelligence can discover and design drug candidates faster. That has already been answered, and the answer is yes. The real question is whether AI can also improve the probability of regulatory success , which is where most promises of disruptive innovation in healthcare historically fail.

Why does this matter to those discussing the PNIRS?

Last week I wrote about Ministerial Decree GM/MS No. 11,921/2026 , which established the National Program for Radical Innovation in Health , and about the apparent difficulty that the Ministry of Health itself demonstrated in defining what, in fact, counts as radical innovation within the program. A sectoral survey conducted with the support of Sindusfarma had already shown that the Brazilian pharmaceutical industry understands radical innovation in a much more restricted way, almost always as a synonym for a new molecule , while the technology sector tends to use the term in a way closer to disruptive innovation , the kind that breaks through a market to create another.

The Insilico case is, in practice, the answer to this false dichotomy. Rentosertib is, at the same time, a new molecule in the strictest sense recognized by the pharmaceutical industry, and the product of a complete technological breakthrough in the discovery process, the meaning that the technology sector attributes to the word "radical ." The two interpretations do not compete with each other. They converge on the same asset. A national program of radical innovation in health needs to reconcile these two understandings and engage with the cross-disciplinary nature of scientific and technological fields.

There is also a second point of connection, perhaps more uncomfortable. Insilico's trajectory up to this point is not the result of a single funded project; it is the result of more than a decade of continuous investment in three specific and clearly defined technological platforms : computational biology , generative chemistry , and clinical outcome prediction , with an almost obsessive focus on solving one problem: making drug discovery faster and more predictable. It is precisely this type of long-term strategic focus that, according to the secretary responsible for the PNIRS (National Program for the Innovation and Research in Health), has not yet been defined for the first Anchor Center of the Brazilian program. Resources of R$ 600 million can, with focus, generate decades of accumulated scientific capacity. Without focus, they tend to be dispersed in experiments that do not add up.

The true rupture: when the boundaries between disciplines disappear

There's a detail in Insilico's trajectory that deserves more attention than it usually receives, and which may be the real "secret" behind this milestone. The classic drug discovery process has always been deeply compartmentalized . Biologists identify and validate therapeutic targets. Chemists design and synthesize candidate molecules. Clinicians design and conduct trials. Statisticians and regulatory specialists interpret results and chart the path to approval. Each stage usually exists within its own institutional culture , with its own language, its own publications, and often a significant loss of context in the transition from one phase to the next.

What Insilico has built is not simply artificial intelligence applied separately to three distinct problems. It's a unique computational architecture where the same database and models traverse biology , chemistry , and clinical design as a continuous flow, not as a conveyor belt of isolated steps. PandaOmics doesn't just point to a target and deliver the result for another team to work on in isolation. It interacts with Chemistry42 to define which chemical structures are viable against that target, and both feed into inClinico , which in turn retroactively informs what types of clinical outcomes and study designs are most likely to confirm the mechanism proposed at the origin, in the choice of the biological target. The boundary between disciplines, which previously required separate teams, deadlines, and translations, dissolves within the same data and model platform.

This is where it's important to differentiate classic radical innovation from the radical innovation we are witnessing now. Historically, a radical advance in healthcare used to originate within a specific discipline : a new class of antibiotics from a novel biological mechanism, a new chemical synthesis route, a new laboratory screening technology. Radicalism was confined within the box of a specialty . What we see now is different: radicalism lies in the very dissolution of the walls between those boxes. It's no longer about having the best biologist, the best chemist, and the best clinician working in sequence. It's about having a platform capable of making all three communicate in real time, within a single learning cycle.

This distinction is not a mere academic detail; it has direct consequences for how an innovation program is designed . An Anchor Center organized as a federation of specialized departments, each in its own discipline, with good administrative coordination among them, is not the same as a platform designed, from data architecture to team governance, so that biology , chemistry , clinical science , and regulation operate as a single, interdependent system . The question that remains for the Arandus Complex and for the next Anchor Centers of the PNIRS is precisely this: does the institutional design being built today favor real transversality between disciplines , or will it reproduce, with a new building and a larger budget, the same logic of juxtaposed departments that has always characterized our science and technology institutions?

From poison to algorithm

It's worth concluding with a reflection I've been developing since I recently revisited the history of COINFAR in a post. COINFAR was a consortium that brought together Biolab , União Química , and Biosintética in 2000, with support from FAPESP , to transform snake venom and venom from other species of Brazilian biodiversity into new medicines. By the standards of that time, it was genuine radical innovation, born from a correct bet on where Brazil's natural comparative advantage lay.

Twenty-six years later, the global frontier of radical innovation in health has shifted not only from physical biodiversity to computational biology . It has also shifted from a model of juxtaposed disciplines to a model of integrated disciplines within a single learning platform. These are two simultaneous transitions, and the second may be more difficult to replicate than the first because it requires rethinking not only which technology to buy or which building to construct, but also how teams of biologists, chemists, clinical practitioners, and regulators are organized , funded , and evaluated within the same institution.

The central question, then, for the current state of the National Program for Innovation in Solid Waste (PNIRS) is: Is Brazil prepared, today, to evaluate or design an innovation program with this cross-cutting nature, capable of connecting different disciplines, potentialities, platforms, and intellectual and academic capacities within a large, cohesive program, and not merely to juxtapose isolated competencies under the same budgetary umbrella?

At the Brazilian Health Innovation Institute - IBIS , we closely monitor this type of boundary, between disciplines, between sectors, between what science already knows how to do and what the market and public policy still need to decide to do with it. If this type of reflection makes sense for your work or your institution, we would be happy to talk.

by Marcio de Paula

Brazilian Health Innovation Institute - IBIS