What a functional biomarker is, and which ones you can compute from your own data

A functional biomarker is a measurement that reflects how well a physiological system is working, rather than how much of one molecule is floating in plasma at one moment. Three shapes: a ratio between two analytes whose balance encodes a state (adiponectin/leptin), a dynamic response to a defined challenge (glucose and insulin after a 75 g oral load), or a rate derived from repeated sampling (glucose variability over 14 days of CGM). Static concentrations are inputs. Functional biomarkers are functions of those inputs, and they are usually the ones that move when something about you changes.

This is a different thing from what the phrase “functional lab testing” often means in consumer marketing, where it describes a large static panel ordered outside of a clinical indication. A 100-analyte panel is still 100 static numbers. Whether it is worth the money depends entirely on what you compute from it and how often you repeat it. The framing we find useful is value as outcome divided by cost, which is the standard argument in laboratory medicine for judging a test by the decision it changes rather than by the number of analytes it reports.¹

The ratios worth computing from a standard blood draw

Most panels already contain the raw material for several functional readouts. You just have to do the arithmetic yourself, because the lab report will not.

Adiponectin/leptin ratio. Both are adipokines, both are cheap add-ons, and their ratio tracks adipose tissue inflammation and dysfunction better than either alone. Frühbeck and colleagues proposed it explicitly as a functional biomarker of adipose tissue inflammation, with ratio thresholds around 1.0 and 0.5 marking increasing degrees of dysfunction.² Units matter and are a common failure mode: adiponectin is reported in µg/mL, leptin in ng/mL. If you divide without checking the units on your PDF, you will be off by 1000. Leptin is also strongly diurnal and rises with sleep deprivation, so draw fasting, in the morning, and note your sleep the night before.

HOMA-IR = fasting insulin (µU/mL) × fasting glucose (mg/dL) / 405. It requires fasting insulin, which most standard panels omit and which you have to request. It is a crude surrogate for a clamp study, but it is repeatable and cheap, and the direction of change over a year is more informative than any single value.

ApoB with LDL-C. ApoB counts atherogenic particles, LDL-C estimates the cholesterol those particles carry. The ratio of the two is a particle-size readout. Discordance between them (high ApoB with unremarkable LDL-C) is the case where the standard lipid panel understates what is going on, and it is a conversation to have with a physician rather than something to act on from a spreadsheet.

Red cell oxidative stress markers. Erythrocytes are a convenient reporter cell for systemic oxidative balance: they are easy to isolate, have no nucleus, and carry measurable malondialdehyde, reduced glutathione, and membrane-bound antioxidant enzyme activity.³ These are functional in the strict sense (enzyme activity, not protein mass), and they are also the most pre-analytically fragile things on this list. Activity assays need the cells processed within hours and often degrade across a freeze-thaw cycle.

Pre-analytics decide whether any of this is real

The largest source of variance in a longitudinal panel is usually not biology. Plasma separation timing, tube type, and centrifugation protocol dominate for a long list of analytes, which is why there is an entire engineering literature on getting plasma off cells faster and more reproducibly.⁴ Practical rules we follow:

• Fix the tube. EDTA plasma and serum are not interchangeable for most proteomic assays. Platelet-derived proteins leak into serum during clotting.
• Fix the time of day and the fasting state. Then keep them fixed forever. A 14-hour fast in March and a 9-hour fast in September are two different experiments.
• Ask for the hemolysis index (H index) with every draw. Hemolysis inflates potassium, LDH, and a large set of proteomic features. If the H index is not reported, you cannot tell a biological signal from a rough venipuncture.
• Record the centrifugation and freezer delay. If a mobile phlebotomist draws at your kitchen table and the tube ships overnight at ambient temperature, your oxidative stress panel is measuring shipping.

Time series beat single draws

CGM is the cheapest functional biomarker generator available, because it samples every five minutes for two weeks. Export from Dexcom Clarity or Libre as CSV. The Dexcom export has a Timestamp (YYYY-MM-DDThh:mm:ss) column and a Glucose Value (mg/dL) column that contains the literal strings Low and High for readings below 40 and above 400. If you call pd.to_numeric without errors="coerce", the column silently becomes object dtype and every downstream statistic is wrong.

For metrics, use the iglu R package rather than rolling your own. It implements the consensus definitions: cv_glu(), mage(), in_range_percent(target_ranges = list(c(70,140))), conga() and j_index(). MAGE in particular has several incompatible published definitions, and hand-implementing it gives you a number nobody else can reproduce. Three metrics carry most of the information: mean glucose, coefficient of variation, and percent of time in a tight range such as 70–140 mg/dL. Run the same standardized meal twice in the same sensor session before comparing anything across sessions, because sensor-to-sensor bias is real and often 5–10 mg/dL.

Where proteomics and RNA-seq earn their place

Affinity proteomics (Olink NPX, SomaScan RFU) gives you 1,000–7,000 features per sample. NPX is a log2 arbitrary unit normalized within a panel and is not comparable across panels or, without bridging samples, across batches. Treat any absolute NPX value as meaningless and every within-subject delta as potentially meaningful. This is the layer where endothelial function becomes visible: circulating adhesion molecules, von Willebrand factor, and endothelial microparticles are the standard readouts of endothelial state, and endothelium is an organ whose function you cannot otherwise sample.⁵ Circulating tight-junction proteins such as claudin-5 and occludin have been studied as readouts of barrier integrity, which is the same idea applied to the blood-brain barrier.⁶ Both are research-grade in an individual, and neither supports a clinical conclusion on its own.

Whole-blood RNA-seq is the other functional layer. Align with STAR or quantify with salmon quant -l A --validateMappings --gcBias, import quant.sf files with tximport, and normalize with DESeq2’s vst() before comparing timepoints. The single most useful derived measure is cell-type composition: whole blood expression is dominated by which leukocytes are present, and deconvolution (CIBERSORTx, or a marker-gene score) turns that confounder into a signal. Leukocyte subset dynamics are themselves functional biomarkers, and the same logic underlies precision diagnostics built on white cell behavior rather than counts alone.⁷ Beyond mRNA, circulating long non-coding RNAs have been associated with left ventricular diastolic function and remodeling, which is an example of a plasma nucleic acid standing in for an organ-level mechanical property.⁸

What we would do

Build a fixed protocol and repeat it. Same fast, same time of day, same tube and same lab, quarterly. Compute a small set of derived measures every time. They are ApoB, adiponectin/leptin, HOMA-IR, CGM CV and time in range, and a leukocyte composition vector. Store them in one tidy table with a draw_id, the pre-analytic metadata, and the raw file paths. Plot each measure against time with its own known assay CV as an error band. Anything that moves less than the assay CV did not move. Anything that moves by two or three times the assay CV across two consecutive draws is worth taking to a physician, with the raw reports, and no interpretation of your own attached.

Questions people also ask

Is functional lab testing worth it? It is worth it if you will repeat the same measurements on a fixed protocol and compute derived quantities from them. A single large panel read once is mostly a source of false positives, since with 100 analytes and 95% reference intervals you expect roughly five out-of-range flags in a healthy person by chance alone.

Are functional blood tests worth it compared to what a physician orders? They cover different ground. Standard clinical panels are optimized for diagnosing disease that is already present. Ratios, challenge tests, and longitudinal series are optimized for detecting change in yourself. Neither replaces the other, and only a clinician can tell you whether an abnormal value needs follow-up.

Is functional testing covered by insurance? Usually not, in the US, when ordered without a clinical indication. Individual analytes on the panel (lipids, HbA1c, TSH) are often covered when a physician orders them for a documented reason. Adipokines, advanced proteomics, and most research assays are not.

How much does functional lab testing cost? Consumer panels of 100+ analytes run a few hundred to about a thousand dollars per round. Adding fasting insulin, adiponectin, leptin, and ApoB to a physician-ordered draw typically costs less than a hundred dollars out of pocket. Affinity proteomics and whole-genome or RNA sequencing are a different order of magnitude.

How often should I repeat? Quarterly is a reasonable default for blood analytes with meaningful biological variation, and two 14-day CGM sessions per year gives you enough data to see a real change in variability. More frequent sampling mostly buys you noise unless you are running a specific intervention with a defined start date.

Footnotes

1. Rossella Tomaiuolo, Giuseppe Banfi. Value is equal to outcome/costs: how to apply to laboratory medicine. Clinical Chemistry and Laboratory Medicine (CCLM), 2025. https://doi.org/10.1515/cclm-2025-0691 ↩

2. Gema Frühbeck, Victoria Catalán, Amaia Rodrı́guez, et al. Adiponectin-leptin Ratio is a Functional Biomarker of Adipose Tissue Inflammation. Nutrients, 2019. https://doi.org/10.3390/nu11020454 ↩

3. Kanti Bhooshan Pandey, Syed Ibrahim Rizvi. BIOMARKERS OF OXIDATIVE STRESS IN RED BLOOD CELLS. Biomedical Papers, 2011. https://doi.org/10.5507/bp.2011.027 ↩

4. Witold Mielczarek, Eleojo A. Obaje, Till T. Bachmann, et al. Microfluidic blood plasma separation for medical diagnostics: is it worth it?. Lab on a Chip, 2016. https://doi.org/10.1039/c6lc00833j ↩

5. Н. В. Гончаров, А. Д. Надеев, Richard O. Jenkins, et al. Markers and Biomarkers of Endothelium: When Something Is Rotten in the State. Oxidative Medicine and Cellular Longevity, 2017. https://doi.org/10.1155/2017/9759735 ↩

6. Erik Axel Andersson, Carina Mallard, C. Joakim Ek. Circulating tight-junction proteins are potential biomarkers for blood–brain barrier function in a model of neonatal hypoxic/ischemic brain injury. Fluids and Barriers of the CNS, 2021. https://doi.org/10.1186/s12987-021-00240-9 ↩

7. 李景荣, Hamed Soleimani Samarkhazan. Decoding Leukocyte Dynamics: Functional Biomarkers and Precision Diagnostics in Parasitic Infections. Acta Parasitologica, 2026. https://doi.org/10.1007/s11686-026-01364-4 ↩

8. David de Gonzalo‐Calvo, Franziska Kenneweg, Claudia Bang, et al. Circulating long-non coding RNAs as biomarkers of left ventricular diastolic function and remodelling in patients with well-controlled type 2 diabetes. Scientific Reports, 2016. https://doi.org/10.1038/srep37354 ↩