Foundation Medicine built its name on comprehensive genomic profiling of tumor tissue — reading a biopsy for the mutations that steer cancer treatment. The harder, larger prize has always been doing the same work from a tube of blood, where there is no surgery, the test can be repeated over time, and the addressable population is far bigger. Patent applications are one of the few forward windows into whether a diagnostics company is actually pushing in that direction, because a publication is a roughly 18-month-delayed snapshot of where the research budget went. Foundation Medicine's filings from the last week of May 2026 point clearly at the blood.

On 28 May 2026, the office published US20260148823A1, "Determining Condition Subtype Based On Fragmentomic Features." The technique it describes — 'fragmentomics' — reads not just the sequence of cell-free DNA in blood but the physical pattern of how that DNA is broken into fragments, which itself carries a signal about the tissue and the disease it came from. The application states the method directly:

An example method includes identifying sequence read data indicating sequences of DNA fragments of a sample obtained from a subject with cancer; and determining, based on the sequence read data, endpoint positions of the DNA fragments with respect to a reference genome.— Determining Condition Subtype Based On Fragmentomic Features, US20260148823A1

The commercial point is the shift in what a blood test can deliver. Detecting that cancer DNA is present in blood is one thing; classifying which subtype of cancer it represents — information that historically required tissue — is a step toward making a liquid biopsy do the work of a tissue biopsy. The filing's CPC classes (G16H 20/10, C12Q 1/6855, G16B 20/20, G16B 40/10) sit at the intersection of clinical informatics, nucleic-acid analysis, and machine learning, which is itself a tell: this is as much an algorithm as an assay.

That distinction carries the commercial weight. The economics of a tissue-based comprehensive genomic profiling test are bounded by the biopsy: a patient has to undergo a procedure, there is often limited tissue, and the test is typically run once. A blood-based test removes the procedure, can be drawn repeatedly to track a cancer over time, and reaches patients who cannot be biopsied at all — a structurally larger market. But blood is a far noisier sample; the fraction of DNA in it that comes from the tumor can be vanishingly small, which is precisely why the value migrates from the wet-lab chemistry toward the computational methods that can extract a reliable signal from that noise. A filing that classifies cancer subtype from the physical fragmentation pattern of blood-borne DNA is an attempt to wring tissue-grade information out of a blood draw — exactly the capability that decides whether a liquid biopsy can substitute for, rather than merely supplement, a tissue test.

The cluster reads the same way

One application is a data point; the surrounding filings are what make it a direction. The same day, 28 May 2026, the office also published US20260148825A1, "Predicting Treatment Efficacy By Analyzing Relative Allelic Expression" — extracting a treatment-response signal from RNA expression data, another layer of clinical inference squeezed from sequencing. Both filings lean on the same computational posture: the value is in what the analysis pulls out of the reads, not in a new reagent.

The pattern continues into June. US20260162765A1, published 11 June 2026, covers a system and method for identifying copy-number alterations — determining tumor purity and ploidy from sequencing-depth signals — which is core machinery for calling a genomic profile from a noisy sample, exactly the noise problem that a blood-based test has to solve. And US20260159894A1, also 11 June, covers IGF1R activating mutations and methods of detecting them — a more classic single-target biomarker filing that shows the company still adding specific actionable alterations to what its profiling looks for. Fragmentomic subtyping, allelic-expression response prediction, copy-number calling, and new actionable mutations: the cluster is a company filing across the full stack of turning sequencing data into clinical answers.

What the signal says, and its limits

Read together, these published applications point to a Foundation Medicine that is investing in pulling more clinical information out of less invasive samples, and doing it through computation — fragment patterns, allelic ratios, statistical copy-number inference. For anyone tracking the diagnostics field, that is a forward read on where the company put R&D in the window these filings reflect: deeper into blood-based and algorithm-driven profiling, not just incrementally more tissue panels. It also indicates the company is layering software-heavy methods (the G16B/G16H informatics classes recur across the cluster) on top of its sequencing base, which is where the durable differentiation in this market increasingly sits.

The caveats are the ones every application read carries. These are publications, not granted patents; claims can narrow before issue, and a filed method is not a launched clinical test. The roughly 18-month publication lag means the cluster reflects earlier spending decisions rather than today's live roadmap. And a diagnostic technique that works in a patent example still has to clear analytical validation and reimbursement before it changes anything commercially. But the direction the filings describe is consistent across late May and into June 2026: more clinical inference, drawn computationally, from blood. The records read as a company steadily moving its profiling franchise from the tumor to the bloodstream — and building the algorithms to make the blood say more.