The first models to learn and extend the central dogma of biology
From DNA, RNA, proteins, peptides, lipids, small molecules, environment, their modifications... to outcome.
Our solution translates omic data to predict phenotype
Genes, transcripts, proteins, and biochemistry each produce an omic layer of biology.
We align all the layers at the machine representation level and traverse them generatively.
We were first to unlock the missing layer
The biochemome contains the small molecules, lipids, and peptides that bridge genes, transcripts, and proteins to phenotype. Biochemical omics has not been broad enough, fast enough, or economical at the scale needed for machine intelligence.
So we pioneered Large Spectral Models (LSMs), the first foundation models to enable direct machine interpretation of spectral data, akin to how language models interpret text, or sequence. Spectral data don't behave like text or image data, and cannot be ingested by the models trained to interpret text and images.
Our current generation LSM was trained on more than 10 billion spectra across millions of molecular structures and biological contexts to make the biochemome machine readable at scale with breadth, speed, and economics on par with sequencing.
Scientific poly-intelligence
It is our process of iterative refinement, where human and machine intelligence collaborate to accomplish together more than each could discover alone. It is an interactive process of hypothesis generation and refinement guided by measured data.
Pyxis reads the layers, interprets them, and predicts the outcomes that matter
Pyxis™ is the co-scientist interface to the collection of Matterworks foundation and task-specific expert models, interacting via common language to accomplish scientific interrogation, critical reasoning, and iteratively challenging & improving results.
A platform and ambition like no other
The most comprehensive training data for biology
Our wedge: we acquired the largest accumulation of unlabeled and expert-curated labeled data for reading the biochemome. We are acquiring the data that extends the layers, aligns them, and connects them to phenotype for generative, predictive translation.
The foundation for foreseeable biology
Language and image models were a start, but they are not equipped for direct machine interpretation of most omic data. We are training the foundation models that ingest raw instrument output to capture its unstructured information content. We were first to do this for biochemical omics, and that is our wedge.
True scientific poly-intelligence
We built Pyxis, the co-scientist interface that allows life science researchers to interact with our models much like they would with their scientific peers and collaborators.
Already in the hands of scientists
We launched PyxisLabs to make Pyxis available to life scientists from discovery through clinical development, diagnostics, and manufacturing. We have started by making de novo chemotyping capabilities for scalable, economical biochemical omics available commercially.
We are just getting started