// mesengenic ai
Hierarchical VAEs and evolutionary density priors for novel biopharma — shortening R&D timelines and expanding out-of-distribution capabilities.
Backed by
// research
Hierarchical Variational Autoencoders and evolutionary datasets for novel biopharma — expanding the search space beyond what natural selection has sampled.
Evolution is a constrained stochastic walk. Traditional high-throughput screening cannot traverse troughs of ruggedness and remains trapped at global maxima. VAEs expand the search space, leveraging latent space navigation to identify the underpinnings of biological fitness. VAE latent kernels serve as a structural compass — bypassing labeling bottlenecks by treating natural evolutionary density as an explicit prior for fitness.
Generative navigation of structural geometries evolution never sampled. To unlock new-to-nature functions we must explore regions of the latent space that are truly new to nature. Mathematical guardrails against fold collapse enable new-to-nature macromolecular engineering with maximal sample efficiency.
To become the true architects of life, we must transition biology from an observational science to an engineering discipline. Epistasis — non-additive frustrated spins in regulatory couplings — is where structure becomes function. Undirected maps capture co-occurrence, not which interaction dominates. Orienting frustrated Jᵢⱼ couplings reveals causal pivot genes and control over lineage commitment — biology as design, not catalogue.
// milestones
Scaled to haematopoietic stem cells — identifying causal pivot genes, the regulators of lineage commitment through causal manifold navigation.
90K+
cells screened
237
gene knockout permutations explored
508M+
interrogated data points
// philosophy
At Mesengenic, we view evolution as a constrained stochastic walk. Natural selection traverses a high-dimensional fitness manifold, but it samples only a fraction of mathematically viable states. We utilise Variational Autoencoders to expand the search space and identify Stable Abiological Manifolds— structural geometries robust enough to sustain novel function, yet entirely unsampled by biological evolution.
Mesengenic exists with two goals: shorten R&D timelines and expand out-of-distribution capabilities. By embedding evolutionary density as an explicit prior and resolving causal structure in frustrated regulatory networks, we navigate fitness landscapes systematically rather than empirically.
The name Mesengenic is not etymology for its own sake. Derived from embryological mechanics — growth from the middle layer — it began with mesenchymal differentiation, where a progenitor branches into lineages through symmetry-breaking.
That metaphor became infrastructure in the haematopoietic stem cell hierarchy. Our HSC work documents a medically urgent, rugged regulatory landscape — decision-forks where frustrated couplings govern lineage commitment — ill-suited to observational science alone. The name records our arc: from mesenchymal inspiration, through symmetry-breaking, to an HSC-anchored causal engine that treats growth-from-the-middle as the governing parameter of latent space exploration.
// contact
Partnerships, research collaborations, or inquiries about our platform — we read every message.