Scientific poly-intelligence in the hands of scientists
Point Pyxis™ to your raw omic data and start a conversation about what analysis you desire, what outcomes you are trying to understand, and what biology you want to discover.
01
Describe the experiment, the nature of the data you are analyzing, and what you want to learn.
02
Pyxis reads the omic data and translates the measurements to what they mean.
03
Explore and refine: work iteratively with Pyxis to improve your experiment, data, and discoveries.
Read about Pyxis knowledge and discoveries
Pyxis leverages Matterworks frontier AI models for predictive biology.
Pyxis co-scientist
Example
You
28-day rat tox study, serum. 12 dosed, 12 vehicle control. Any sign of liver stress, and through what mechanism?
Pyxis
The dosed animals separate cleanly on a bile-acid signature. Conjugated bile acids (taurocholate, glycocholate) rise 3 to 4 fold while lysophosphatidylcholines drop, a pattern consistent with early cholestatic liver stress, before ALT moves. Three animals sit closest to the control group.
Rank animals by severity
Show the pathway
Compare to a known cholestatic
Ask a follow-up…
Your data and discoveries are yours
Private data relationship
Under our commercial licenses, we do not train on your data. Providing information about your samples in your private data environment allows Pyxis to do more, interpret more, and provide better poly-intelligence for your benefit, using your proprietary context.
Models tuned to you
Enterprise programs run on models fine-tuned to your biology alone. They never serve another customer.
Every layer of biology provides a slice of the central dogma that builds from genotype to phenotype
The original central dogma of biology started from DNA, and built through RNA to proteins. Translating to phenotype requires extending the chain to include the biochemome, the overlooked layer. We solved that first.
Phenotype
The outcome itself. Efficacy, toxicity, titer, or disease state, and the thing every program is finally trying to predict.
Biochemome
Small molecules, lipids, and peptides reflect the actual biochemical state, and sit closest to phenotype. Without this layer, the translational signal goes dark. We started by making this layer visible, interpretable, and scalable, on par with sequencing.
Into the biochemome
>1,000x
small molecules, lipids, and peptides per protein
Proteome
What machinery is present. Closer to function, and still upstream of whether that machinery is actually running.
100x
proteins per transcript
Transcriptome
What is being asked for. Which genes are being transcribed right now, though transcription does not reliably carry through into protein.
10x
transcripts per gene
Genome
What could happen. Stable, comprehensive, and inexpensive to read, and separated from the outcome by everything regulation does downstream.
Where this leads
As the models learn the alignment between layers, the mapping from biology to outcome becomes something you can run in either direction.
In development
Predict forward
From what you measured to what will happen. Read a sample now, and Pyxis returns the outcome it points to, ahead of the animal study, the cohort, or the batch record that would confirm it later.
In development
Design backward
From the outcome you want to the biology that produces it. Name the phenotype you are aiming at, and Pyxis works back to the mechanism, and to the molecules that would move it.
Enterprise programs
Close the loop
Every result sharpens the model working on your programs. Fine-tuning happens inside your private relationship, so what you learn compounds for you and stays with you.
Engage with Pyxis on a new experiment or discover the novel biology hidden in data you already have.