Evolution’s Shortcuts, Engineered: Enveda’s Edge in Drug Discovery

This post builds on our earlier blog here.

‍

We are delighted to build on our partnership with Enveda, leading their Series D financing as they continue to redefine small molecule drug discovery. The company has made tremendous progress over the last 12 months, including its first drug candidate progressing to clinical trials and an expansion of the development pipeline to include several more exciting molecules. These early milestones support the hypothesis of what we called “Enveda’s meta-experiment”, opening new sources of novel drugs and a more efficient drug discovery process.

‍

“Natural-like” starting points can tilt odds toward safer, more effective drugs

Imagine proteins as locks and small molecules as keys. Nature has spent billions of years “cutting” keys (metabolites, defense chemicals) that fit biological locks. When you start with keys from that drawer, you’re more likely to find ones that turn the right locks.

‍

A consistent finding across decades of approvals is the outsize role played by natural products and their derivatives. Newman & Cragg’s comprehensive census (1981–2019) shows nature‑derived scaffolds contributing across oncology, anti‑infectives, cardiometabolic and more. Many approved drugs are closer to human metabolites than random library compounds, which can help with tissue access and predictable characteristics. None of this means “natural = safe” but it does mean nature‑like chemotypes often give a head start on balancing potency, exposure, and off‑target risk.

‍

However, most biotechs don’t start from raw natural molecules because it’s slow, expensive, and unpredictable path to finding tractable bioactive lead molecules. Making analogues to improve properties can require specialised, multi-step chemistry as there are no pre-validated routes. What is extracted over long periods of experimentation, might turn out to be an already known compound. In general, this leads to long, uncertain timelines and significant spend before you generate meaningful data. Instead, most biotechs use synthetic libraries that pre‑plan synthesis using validated reactions. This lowers upfront capital needs and lets small teams run multiple shots on goal in parallel.

‍

The search space for new drugs is so vast, and conventional approaches miss most of it‍

For discovery teams, the practical question isn’t “what exists?” but “what can we actually search?” This practicality comes at a cost.

‍‍

Modern chemistry’s “universe” is enormous, with the number of theoretically drug-like molecules estimated anywhere from ~10³³ to ~10⁶⁰. Even though the theoretical universe is huge, the libraries we can realistically screen are far smaller. The largest ones are the make‑on‑demand spaces that pre‑plan synthesis using validated reactions such as Enamine REAL Space, which (as of 2025) comprises ~76.9 billion synthetically accessible molecules.

‍

Critically, many of these molecules can be made and shipped within a few weeks because routes are pre‑validated. However, this advantage comes with a key caveat, even at ~77B scale, make-on-demand libraries are shaped by the building blocks and reaction rules they’re built from. As libraries shift from in-stock to make-on-demand, their similarity to “bio-like” molecules (drugs, metabolites, natural products) drops sharply. In practice, this means natural-like chemotypes needing bespoke, multistep synthesis are underrepresented.

‍

How Enveda incorporates AI into drug discovery to drive efficiency‍

While there is potentially a more fertile space available for drug discovery, accessing this in an efficient manner is key to a repeatable drug discovery process. How can we turn slow & unpredictable to fast & predictable?‍

Enter Enveda’s insight on the problem, rather than brute‑forcing all corners of synthetic space, start where biology has already explored and make it computable. Enveda first assembled a knowledge base linking thousands of plants to diseases and symptoms, then paired high‑throughput mass spectrometry with AI so the platform can “read” the spectral fingerprints of molecules derived from these sources at scale. Their foundation model, PRISM, was trained on ~1.2 billion tandem mass spectra, enabling faster identification and triage of promising molecules.

‍

Mass spectrometry turns a molecule into a barcode. Traditionally, reading that barcode for an unknown molecule required isolating it and running time‑consuming experiments. Enveda’s PRISM is a foundation model that learns the “grammar” of spectra via self‑supervision (similar to how language models learn by predicting masked words). Trained on ~1.2B spectra from public repositories and Enveda’s automated lab, PRISM boosts downstream tasks like structure/property prediction and “closest match” retrieval. This turns weeks of bench work into computation measured in seconds.

‍

This AI layer sits inside a broader loop:

(1) Organize with the plant–disease knowledge base to prioritize promising samples.

(2) Translate by profiling mixtures and using PRISM to infer structures/properties for thousands of candidates in parallel; and

(3) Apply with a fully automated lab to test and triage rapidly.

‍

Enveda attributes resulting efficiency gains to this loop, citing ~4× faster to development candidate, ~75% fewer analogues per DC, and an ~11× higher per‑scaffold success rate versus industry baselines.

‍

Unlocking novel biology and broader disease reach

What if weeks of extraction, purification, and structure-solving before you even know what you have, often ends up in rediscovering a known compound? Are there enough novel drugs to be found in this space?

‍‍

Enveda’s early pipeline suggests that the answer is a resounding “Yes!”. The input chemistry for Enveda’s drug pipeline is different and because the platform can read tens of thousands of molecules per run, the hits often sit outside the usual synthetic playbooks. Recently, ENV‑294, a first‑in‑class oral anti‑inflammatory candidate, cleared Phase 1 with a favorable safety profile and moved into a Phase 1b study for atopic dermatitis. Beyond dermatology, Enveda pipeline consists of multiple small‑molecule programs across multiple therapeutic segments.

‍

Two testable ideas underlie the platform. First, that “natural-like” chemistry will translate into better safety/ADME and target selectivity on average, a thesis supported by retrospective analyses but still subject to indication-by-indication proof. Second, that foundation-model gains in reading mass spectra at scale will keep compounding (more spectra → better embeddings → fewer wet-lab dead-ends). If both continue to hold, Enveda’s mix of ethnobotanical priors, spectral AI, and causal biological reasoning could reset the cost/time curve for small-molecule discovery while surfacing pathways and indications industry screens often overlook.

‍

We believe Enveda’s approach is a credible way to expand hit-finding and compress the path to the clinic while unlocking novel biological pathways and indications. In five years, the company has created a pipeline of drug candidates which target novel biology, in areas large unmet need. This pace of drug discovery is unprecedented, and the breadth of the pipeline indicates the process is repeatable. Even at success rates comparable to traditional biotechs, this translates to a generational pharma company. Further, we expect the proposition that starting from “natural-like” chemistry will translate into better treatments, will play out in the clinical trials. In combination, this is a significant step forward towards a more predictable, technology-enabled, drug discovery process and opens new sources of medicines for patients worldwide. Early clinical progress and platform metrics are consistent with this thesis, and the best is yet to come, in our view.