deSIPHR:
proteomic data infrastructure at semiconductor scale

Microscope examining a colorful nanophotonic wafer with a red laser light.

The proteomics challenge

Proteomics has always faced a fundamental constraint: the molecules that execute biology are extraordinarily diverse, dynamic, and difficult to read at scale.

The current workhorse — mass spectrometry — has transformed our understanding of the proteome. But most workflows identify proteins indirectly, by fragmenting molecules and interpreting the resulting spectra, with the aid of reference databases. Other approaches rely on fluorescent labels or antibody-based affinity methods, trading breadth for sensitivity and targeted measurement.

The result is a persistent measurement gap. The field has spent decades generating an increasingly detailed map of the best-lit corners of the proteome, while much of biology’s functional molecular layer remains hidden.

This isn't a sensitivity problem. It's a category problem. Biology needs methods capable of reading its functional layer directly, with the resolution to distinguish molecular identity and state and the scale to capture biological complexity.

And identity is only the beginning: understanding biology ultimately requires measuring not only which molecules are present, but how they interact and how those interactions shape function.

Pumpkinseed is building the technology to read that layer.

How deSIPHR works

deSIPHR is Pumpkinseed's proprietary nanophotonic chip platform for measuring protein biology. With over 100 billion sensors per wafer, it is designed to interrogate proteins across multiple dimensions - from molecular identity and modification state to molecular interactions.

deSIPHR reads molecular building blocks one at a time using Raman spectroscopy. No reference proteins. No elaborate sample prep. The result is direct, high-resolution proteomic data at a scale and fidelity that simply hasn't existed before.

What is Raman spectroscopy? Raman spectroscopy reads the molecular vibrations of individual molecules, revealing rich information about their identity, structure, and interactions. This is physics reading biology in the most literal sense.

The chip itself is fabricated using semiconductor manufacturing — the same processes that produce modern microchips — which means it scales the way silicon scales. More sensors. Higher throughput. Lower cost per data point over time.

Schematic representation of how the deSIPHR platform works.
Schematic representation of how the deSIPHR platform works.
Close-up of multiple colorful nanophotonic wafers, showing bright, reflective surface with intricate circuitry and gold contacts, inside a plastic protective packaging.

A technology built for this moment

  • De novo, by design. deSIPHR sequences proteins without prior knowledge of what's in the sample. No reference database required. No hypothesis about what to look for. This is the fundamental shift from confirmation to discovery.

  • Single-molecule resolution. The platform reads proteins at the level of individual molecules, capturing the full complexity of the proteome, including low-abundance proteins, post-translational modifications, and proteoforms that existing methods miss entirely.

  • Scale that matters. With over 100 billion sensors per wafer, deSIPHR generates proteomic data at a throughput that makes meaningful datasets possible. Not individual experiments, but the kind of rich, comprehensive coverage that AI biology and drug discovery actually need.

  • More than identity. Biology is defined not only by which proteins are present, but by what they interact with and how those interactions change function. deSIPHR’s vibrational readout creates the potential to connect molecular identity with interactions on the same massively parallel measurement architecture - generating data that links what proteins are to what they do.

From signal to dataset

Every deSIPHR experiment generates structured, high-dimensional protein data: identify, sequence, modification state, abundance, and interactions.

Our goal is not simply to produce more proteomic measurements. It is to build datasets that connect what proteins are to what they do - identity to interaction, interaction to functional state, and functional state to biological response.

These datasets don’t already exist at the scale BioAI needs. They can’t be scraped from the internet. They have to be experimentally generated, and deSIPHR is being built as the infrastructure to generate them. The result is the high-resolution, linked biological data needed to train and validate predictive models of biology.