Data
The full data, sample by sample
Every blood sample behind the headline numbers, with how events were counted and what the data can and cannot show. These are exploratory results from earlier studies; our NIH Phase I plan repeats them under controlled conditions.
How we count
Reading the numbers
After capture, cells are stained with a tumor marker (EpCAM or cytokeratin) and the white-cell marker CD45, then read one by one on a flow cytometer. An event counts as tumor-compatible when it is bright for the tumor marker, dim for CD45 and large enough to be a cell, using fixed numerical thresholds applied to the raw instrument files. The same rule is applied to spiked and unspiked samples.
Because samples differ in how many total events were read, we compare rates per 10,000 events, not raw counts. "Spiked" samples had about 100 cultured cancer cells added to healthy blood by dilution, so the exact input is not known and we do not report a recovery percentage.
Spike test 1 · lung
A549 lung cancer cells vs. unspiked blood processed the same way
About 100 A549 lung cancer cells were added to 7.5 mL of healthy blood and processed through the charge-tuned microtubes. Two 7.5 mL blood samples with nothing added went through exactly the same protocol, one from the same donor. This is our most closely matched comparison. Cells were stained for cytokeratin after fixation.
tumor-compatible events in the spiked sample
events in the two unspiked controls
higher rate than the same donor's unspiked blood
higher rate than the second donor
Tumor-compatible events per 10,000
Same protocol for all three samples; same instrument settings.
| Sample | Cells added | Events | Total read | Per 10k | Spike ÷ control |
|---|---|---|---|---|---|
| Lung spike (product) | ~100 A549 cells | 158 | 303,159 | 5.21 | |
| Control A (same donor, 2 wk later) | none | 20 | 320,946 | 0.62 | 8.4× (95% CI 5.2–14.1) |
| Control B (different donor) | none | 24 | 555,848 | 0.43 | 12.1× (95% CI 7.8–19.4) |
Rule: cytokeratin > 200, CD45 < 100, size > 20,000. Exact conditional binomial 95% confidence intervals.
Does the rule pick out tumor cells?
We applied the same rule, unchanged, to separate staining tests on pure cell populations that were not used to set it.
| Pure sample | Events | Pass the rule |
|---|---|---|
| A549 lung, cytokeratin-stained | 2,606 | 74% |
| COLO205 colon, cytokeratin-stained | 1,330 | 72% |
| Pancreatic line, cytokeratin-stained | 9,321 / 10,987 | 79% / 68% |
| Unstained tumor cells | various | 0–0.3% |
| White cells, cytokeratin + CD45 | 1,976 | 0.4% |
Limits
- The thresholds were chosen using this experiment, so a new, independent spike is needed to confirm them.
- The colon rule does not carry over to this fixed cytokeratin stain (it gives 1.5×, not distinguishable from background), which is why each stain has its own rule and the two experiments are never pooled.
- The excess (~138 events) is larger than the nominal ~100-cell input, which was made by dilution; clumps, fragments or input error may contribute. No recovery is inferred.
- One spike experiment; individual events are not separately confirmed as A549 cells.
Spike test 2 · colon
COLO205 colon cancer cells vs. 15 healthy donors
About 100 COLO205 cells were added to about 7.5 mL of healthy blood and processed through 26 microtubes. The recovered cells were compared with 15 healthy-donor samples from an earlier series that also went through the Bioforce workflow, under related but not identical settings, and were read on the same instrument with identical settings. Because the controls were not run side by side, this comparison is less direct than the lung test.
tumor-compatible events in the spiked sample (38 after removing saturated-signal events)
such events in all 15 donors combined (3.3 million events read)
higher rate in the spike than in the pooled donors (15.2× after the saturation filter)
donors with a lower rate than the spike
Tumor-compatible events per 10,000, by sample
Spiked sample in orange; each gray bar is one healthy donor.
| Sample | Reads | Events | Clean | Total read | Per 10k | Clean/10k | Spike ÷ donor |
|---|---|---|---|---|---|---|---|
| Colon spike (product) | 42 | 79 | 38 | 551,953 | 1.431 | 0.688 | |
| Donor 45 | 8 | 3 | 1 | 44,986 | 0.667 | 0.222 | 2.1× |
| Donor 46 | 8 | 0 | 0 | 87,369 | 0.000 | 0.000 | – |
| Donor 47 | 10 | 2 | 1 | 103,365 | 0.193 | 0.097 | 7.4× |
| Donor 48 | 8 | 1 | 0 | 54,145 | 0.185 | 0.000 | 7.7× |
| Donor 49 | 10 | 2 | 2 | 157,022 | 0.127 | 0.127 | 11.2× |
| Donor 50 | 8 | 1 | 1 | 154,298 | 0.065 | 0.065 | 22.1× |
| Donor 51 | 9 | 2 | 1 | 149,754 | 0.134 | 0.067 | 10.7× |
| Donor 52 | 9 | 1 | 0 | 104,064 | 0.096 | 0.000 | 14.9× |
| Donor 53 | 9 | 2 | 0 | 95,505 | 0.209 | 0.000 | 6.8× |
| Donor 54 | 9 | 0 | 0 | 85,634 | 0.000 | 0.000 | – |
| Donor 55 | 9 | 0 | 0 | 71,895 | 0.000 | 0.000 | – |
| Donor 56 | 12 | 1 | 1 | 502,030 | 0.020 | 0.020 | 71.9× |
| Donor 57 | 24 | 5 | 5 | 1,251,720 | 0.040 | 0.040 | 35.8× |
| Donor 58 | 9 | 2 | 2 | 190,611 | 0.105 | 0.105 | 13.6× |
| Donor 60 | 10 | 1 | 1 | 265,764 | 0.038 | 0.038 | 38.0× |
| 15 donors, pooled | 152 | 23 | 15 | 3,318,162 | 0.069 | 0.045 | 20.6× |
| 15 donors, average | 0.125 | 0.052 | 11.4× |
"Clean" excludes events whose side-scatter signal hit the top of the instrument range. Rule: EpCAM signal > 1000, CD45 < 150, size > 5000 (raw linear units). The noisiest donor (45) still sits 2.1× below the spike.
Limits
- The donors were processed months earlier under related but not identical settings (different buffer and wash conditions), so this is not a matched spiked-versus-unspiked experiment.
- The stain rule identifies tumor-like cells; it does not prove each event is a COLO205 cell. A few donors also had rare qualifying events.
- About 100 cells were added by dilution, not counted one by one, so 79 events is not "79% recovery."
- Count history: the original 2021 hand-drawn count was 53, and a strict-rule figure of 10 was later cited for the first of the 26 tubes only. Applying the strict rule to all 26 tubes gives 79.
Patient blood · prostate cancer
Tumor cells from seven patients with advanced prostate cancer
Processed at Cornell in 2017 with a related charged-surface method. Captured cells were fixed, stained (EpCAM, cytokeratin, PSMA, DAPI, CD45) and classified with Weill Cornell collaborators.
Tumor cells counted per 3.5 mL of blood
Includes EpCAM-negative tumor cells. Three patients had 150–310 cells; four had 1–11.
Pure-population capture
At one operating point, separate runs of pure cell populations captured:
of COLO205 tumor cells (± 4.0%)
of white cells (± 1.9%)
A 51.5-fold ratio of means. Mixed whole-blood samples are harder, which is why purity is the main engineering target.
Mixed-cell purity (April 2020, preliminary): in a buffer run with cultured cancer cells mixed into white blood cells, about 40% of captured cells were tumor cells (two samples). Purity in whole blood is much lower, and raising it is the main target of our NIH Phase I work.
What comes next
How Phase I tightens this
- Count-verified spikes instead of dilution, so recovery can be measured.
- Unspiked controls run side by side, same day, same donor, same protocol.
- Analysis rules locked before the confirmation runs, with every raw file kept.
- Viability and culture of the recovered cells, measured directly.
Full computational reports (parser, per-acquisition settings check and file hashes) are available to reviewers and partners on request.
Questions about the data?
We are happy to walk through the raw files with reviewers, partners and customers.
For research use only. Not for use in diagnostic procedures.