Variational AI’s Enki 4 Jumps Into Advanced Drugs, Rattling Rivals

TL;DR

• Variational AI launched Enki 4, a major upgrade to its generative AI platform for small-molecule drug discovery, now covering 760 drug targets — a 28% jump from the previous version.
• The platform now supports degraders, PROTACs, molecular glues, and novel payloads for antibody-drug conjugates, expanding beyond traditional small molecules into cutting-edge therapeutic modalities.
• Enki 4 features re-architected algorithms that deliver faster performance while generating more potent, selective, and synthesizable drug candidates for biopharma partners.
• The upgrade sharpens Variational AI’s competitive edge against rivals like Insilico Medicine and Recursion Pharmaceuticals in the rapidly growing AI drug design market.

Variational AI Expands Enki 4 to 760 Drug Targets

Variational AI announced Enki 4, a significant upgrade to its generative AI platform for small-molecule drug discovery. The new version expands target coverage to 760 drug targets — a 28% increase over its predecessor — and adds support for degraders, PROTACs, molecular glues, and novel payloads for antibody-drug conjugates.

The platform now operates faster while delivering improved algorithmic performance across a broader range of therapeutic modalities. Ali Saberali, Co-Founder and Head of Platform at Variational AI, said the release represents a fundamental overhaul. “Enki 4 is a massive step forward for our platform. We’ve re-architected Enki and improved the underlying algorithms to expand target and modality coverage, while operating faster to deliver even better performance for our partners.”

The upgrade aims to accelerate biopharma R&D pipelines by enabling rapid generation of novel molecules that are potent, selective, and synthesizable. The addition of PROTAC and degrader support is particularly notable — these modalities represent some of the most promising frontiers in targeted protein degradation, a therapeutic approach that’s attracted billions in investment over the past few years.

Why Enki 4’s Modality Expansion Reshapes Drug Design Timelines

This isn’t just a version bump. It’s a bet that generative AI can crack therapeutic modalities that have historically required years of iterative chemistry.

Traditional small-molecule drug discovery is already a slog — it takes an average of 10-15 years and reportedly over $2 billion to bring a single drug to market. Now add PROTACs and molecular glues, which are structurally complex and require precise optimization to degrade target proteins without off-target effects. The design space explodes. The failure rate climbs even higher.

Enki 4’s expansion to 760 targets — up 28% — matters because it widens the aperture on what’s druggable. Many disease-relevant proteins have been considered “undruggable” because they lack convenient binding pockets for traditional small molecules. But degraders and PROTACs sidestep that constraint by hijacking the cell’s own protein disposal machinery. If Variational AI can generate optimized degraders computationally, it collapses what used to take medicinal chemistry teams months or years into days or weeks.

And that’s the real edge here — speed married to specificity. Generative models like Enki don’t just spit out random molecules and hope for the best. They’re trained to optimize for multiple properties simultaneously: potency, selectivity, synthesizability, and now modality-specific constraints like linker chemistry for PROTACs or E3 ligase engagement for degraders. It’s like asking a design tool to solve a Rubik’s cube where every face represents a different drug property, and it has to align all of them at once.

I’ve watched generative AI in drug discovery evolve from a curiosity to a competitive necessity over the past five years. What strikes me about Enki 4 is the modality breadth. Most platforms still focus narrowly on traditional small molecules. Variational AI is making a calculated move to own the next generation of therapeutic formats before they become commoditized.

But there’s a counterargument worth considering: does broader target coverage dilute performance? Generative models are only as good as their training data, and PROTACs are still a relatively young field with limited experimental datasets. If Enki 4 is stretching across 760 targets and multiple modalities, is it sacrificing depth for breadth? The company’s claim of “improved algorithmic performance” suggests they’ve addressed this, but the proof will be in the clinical candidates that emerge from partner pipelines over the next 18-24 months.

How Variational AI Stacks Up Against Insilico and Recursion

Variational AI isn’t operating in a vacuum. The AI drug design market is crowded with well-funded competitors, and the stakes are enormous.

Insilico Medicine has raised hundreds of millions and already advanced AI-designed molecules into clinical trials. Recursion Pharmaceuticals went public and is betting big on combining wet-lab automation with AI-driven phenotypic screening. Both companies have built massive datasets and computational infrastructure. Both are racing to prove that AI can actually deliver FDA-approved drugs, not just interesting molecules.

Variational AI’s edge — at least on paper — is its foundation model approach. The company was founded by machine learning researchers from MIT, Caltech, Google Research, Microsoft Research, and D-Wave. That pedigree shows in Enki’s architecture, which uses state-of-the-art generative models to optimize molecules across multiple objectives simultaneously.

The PROTAC and degrader support in Enki 4 is a differentiator. Neither Insilico nor Recursion has publicly emphasized these modalities to the same degree. If Variational AI can demonstrate superior performance in targeted protein degradation — a field where experimental iteration is especially expensive — it carves out a defensible niche.

But the competitive threat cuts both ways. Recursion’s scale gives it an advantage in data generation. Insilico’s clinical progress gives it credibility with pharma partners. Variational AI needs to translate Enki 4’s expanded capabilities into partnership wins and, eventually, clinical candidates. Algorithmic sophistication matters, but pharma ultimately pays for molecules that work in humans.

Generative AI’s Growing Role in Biopharma Pipelines

Zoom out, and Enki 4 is another data point in a larger trend: generative AI is moving from proof-of-concept to production in drug discovery.

Five years ago, most biopharma companies treated AI as a side project — interesting, maybe useful, but not core to R&D strategy. Today, nearly every major pharma has inked partnerships with AI-driven biotech companies or built internal AI capabilities. The shift happened because the economics are undeniable. If you can cut early-stage discovery timelines by 30-50%, you compress cash burn and get to clinical readouts faster. That changes the risk-reward calculus for the entire pipeline.

Variational AI’s focus on synthesizability is particularly smart. One of the early criticisms of generative models in chemistry was that they’d propose molecules that looked great in silico but were impossible or prohibitively expensive to make in the lab. Enki’s emphasis on generating synthesizable candidates addresses that head-on. It’s not enough to design a molecule with perfect binding affinity if your medicinal chemists can’t actually make it.

The addition of ADC payload support is also forward-looking. Antibody-drug conjugates are one of the hottest areas in oncology, and the payload — the toxic molecule attached to the antibody — is critical to efficacy and safety. If Enki 4 can accelerate payload optimization, it taps into a multi-billion-dollar market where even incremental improvements in therapeutic index translate to blockbuster potential.

What Variational AI Needs to Prove Next

The real test for Enki 4 isn’t the feature list. It’s whether those 760 targets and expanded modalities translate into clinical candidates that outperform traditionally designed molecules. Variational AI needs to show that its re-architected algorithms don’t just work in theory — they work in cells, in animals, and eventually in patients.

Watch for partnership announcements over the next six months. Pharma companies are hungry for AI-designed PROTACs and degraders, but they’re also cautious. If Variational AI can land deals with top-tier partners — think Pfizer, Roche, or Novartis — it signals that Enki 4’s capabilities are passing internal due diligence. Those partnerships often come with upfront payments, milestone fees, and royalties that can fund further platform development.

Also watch for clinical milestones from existing partners. Variational AI hasn’t disclosed how many molecules designed by earlier Enki versions are in preclinical or clinical development, but that pipeline is the ultimate validation. If an Enki-designed molecule hits a Phase I safety milestone or shows efficacy in Phase II, it’s a proof point that reverberates across the industry. It tells every other pharma company that generative AI isn’t just hype — it’s a competitive advantage they can’t afford to ignore.

Source: AFP