When a DNA Peak Runs Off the Scale Rebuilding a Saturated Signal From the Detector Next Door

This application is filed under the title “Pattern Recognition System” and carries an abstract describing a broad adaptive expert system for interpreting DNA data, but the claims are much narrower than either suggests. Claim 1 is a piece of arithmetic: when a fluorescence peak is so bright that one detector maxes out and flatlines, use a second detector that did not max out, work out the fixed ratio between the two in the part of the peak where both are still behaving, and use that ratio to reconstruct the height the first detector would have reported. The applicant is Ande Corporation, the Massachusetts maker of the ANDE Rapid DNA instruments used at booking stations and disaster sites, and the named inventors are Eugene Tan, Hua Jiang and Richard F. Selden. It is a divisional of Australian application 2024203902, which was itself divided from AU 2018300157.

The Problem

Human identification by DNA rests on short tandem repeats, short sequences of two to six bases that repeat a variable number of times at a given position in the genome. The specification notes there are several hundred thousand such loci in the human genome, and that a person’s repeat counts across a couple of dozen of them are distinctive enough to identify them. The process has three steps: extract and purify the DNA, amplify the chosen loci by PCR using fluorescently labelled primers, then separate the amplified fragments by electrophoresis and detect the fluorescence as the fragments pass a window. The output is an electropherogram, a trace of coloured peaks, and reading which peaks are real alleles is the allele calling step.

Allele calling is where the trouble is. The specification is blunt about how much DNA you need to get a clean trace. Too little and you get peak height imbalance, allele drop-out and drop-in. Too much and you get increased stutter, inflated incomplete non-template addition peaks, pull-up or bleed-through between colour channels, extra wide peaks, and what the document calls peak “flat-topping”, where a high signal is simply cut off at the upper limit of the detection system. That upper limit is a hard number in this instrument: the specification states that in the ANDE system the detector returns a maximum value of 262,144, and any peak that would have gone higher reports that value and nothing more.

This is why forensic laboratories quantify DNA before amplifying, a step the FBI has required for profiles that are to be searched against federal and state databases. Quantitation is a way of keeping the sample inside the instrument’s dynamic range. It costs time and it costs sample.

The other half of the problem is who does the reading. The specification quotes the FBI’s position that there are no expert systems approved for use on casework, as distinct from reference, samples, and describes the conventional arrangement in detail: the expert system applies fixed rules, flags anomalies, and the human analyst then manually edits rules and parameters, overrides individual peaks, and makes the pass or fail decision. Other analysts review that. The document says this reliance on human interpretation is “highly subjective and can vary from analysts to analyst and, for a given analyst, from day to day”, and that reprocessing the sample may occur two, three, five, ten or more times. That is the contentious part of automating the call. A profile that goes to court has traditionally had a named person prepared to defend every peak in it.

The specification gives a concrete illustration of what over-range data does to a conventional expert system. In its Figure 3, a high DNA content sample has average heterozygote peak heights above 10,000 RFU, wide peaks that fail the width threshold and go unlabelled, incomplete non-template addition peaks so tall they are labelled as real alleles, and two loci showing three alleles each, which made the standard system treat the sample as a mixture and flag every allele in red. The specification then makes the point that matters: all of those red boxes were on correct peaks, and the profile actually contained enough information for a random match probability better than one in a sextillion. The standard system failed the sample anyway.

What This Invention Does

Claim 1 is an adaptive expert system comprising a computing device with memory and a processor, executing instructions to do five things. Receive sample data from an optical detection system having at least two detectors or detector elements, those detectors being configured to collect fluorescence from dye-labelled DNA and turn it into an optical signal. Subtract a baseline from that signal. Then, for at least one peak where at least one detector saturates and at least one does not, calculate the signal strength ratio between each saturated detector and each unsaturated detector, measured only in the portion of the peak where neither is saturated. Finally, calculate the signal strength of the saturated detector across the saturated portion of the peak by multiplying the unsaturated detector’s signal by that ratio.

The logic is straightforward once stated plainly. A dye’s emission is picked up by more than one detector, each tuned to a different wavelength, at different intensities. One of them is the primary for that dye and goes off scale first. The others see a weaker version of exactly the same peak and stay within range. Because both detectors are watching the same physical event, the proportion between them is fixed, and it can be measured on the flanks of the peak before the primary hits its ceiling. Knowing that proportion, the flat top can be replaced with the shape the primary detector would have produced if it had more headroom. The specification describes the result as an optical signal “that is no longer saturated” and quantified “as if the optical signals of the primary detector were not saturated”.

The body of the specification sets this out as a “peak quantification module” at paragraph [00246], with a four step algorithm: detect and store the saturated and unsaturated portions of each detector signal, identify the secondary detector signals for the same peak, calculate the primary over secondary ratio across the part of the peak where the primary is still linear, and then use that ratio in the saturated region. One thing worth flagging for anyone working from this document: the body says to do the last step by “dividing the signal strength of the secondary peak by the Primary/Secondary detector signal strength ratio”, while claim 1 says to multiply. Given the ratio is defined as primary over secondary, the claim is arithmetically right and the body reads like a slip.

The dependent claims are unusually spare. Claims 2 through 11 are nothing but a ladder of detector counts: at least 4, then 5, 6, 8, 12, 16, 32, 64, 128, and finally at least 256 detectors or detector elements. There are no dependent claims covering thresholds, rules, iteration or anything else. The claim set ends at claim 11.

That matters because the rest of the 211 page specification is about something else. The document spends most of its length on what it calls an Adaptive Expert System, which sorts profiles into signal strength categories and then iterates through pre-built parameter sets to find the one that calls the most loci without flipping a homozygote into a heterozygote. Example 8 describes nineteen parameter sets for low signal samples, generated by walking the heterozygote peak height threshold from 250 down to 100 RFU, the heterozygote peak height ratio from 0.35 down to 0.15, the homozygote threshold from 300 down to 120 RFU, and the hemizygote minimum from 200 down to 80 RFU, plus three parameter sets for high signal samples and a rule that fires when the incomplete non-template addition ratio at the amelogenin locus reaches 0.4. Example 4 reports that median heterozygote peak heights across the working dataset ran from 1,580 to 49,220 RFU with a median of 8,180 RFU, and uses that median as the dividing line between low and high signal samples. None of that is claimed in this divisional. Under Australian practice the parent and its other divisionals hold those claims, and this document says as much in its opening paragraphs.

Key Features

• Cross detector ratio recovery. The core step is measuring the fixed proportion between a saturated detector and an unsaturated one in the region where both are still linear, then using it to reconstruct the clipped region.
• A ratio measured only where both detectors are honest. Claim 1 is specific that the ratio is calculated “in a portion of the peak where both detectors or detector elements are not saturated”, which is what keeps the reconstruction from being contaminated by the clipped data it is trying to replace.
• Baseline subtraction before the ratio is taken. The claim puts baseline subtraction ahead of the calculation, so the ratio is between two corrected signals rather than two signals sitting on different offsets.
• A hard, instrument specific saturation value. The specification defines saturation as the detector returning its maximum value, and states that in the ANDE system this value is 262,144, which is two to the power of eighteen.
• Scaling to many detector elements. The only dependent claims are a ladder of detector counts running from at least 4 up to at least 256, aimed at linear array detectors rather than the four or five colour channels of a conventional sequencer.
• Dynamic range extended without touching the chemistry. Nothing in the claim changes the extraction, amplification or separation. The fix is applied to data that has already been collected, which is the point when the sample may be gone.

Who Is Behind It

Ande Corporation is a private company based in Waltham, Massachusetts, and its Rapid DNA instruments are designed to take a swab and return an STR profile in about ninety minutes in a police station, a military forward base or a disaster mortuary, operated by someone with no laboratory training. The specification describes the ANDE 6C system as a ruggedised, fully integrated instrument using a single disposable microfluidic chip that does extraction, purification, PCR, electrophoresis, fluorescence detection and on-board analysis, and notes that in May 2018 it became the first rapid DNA system to receive the FBI’s National DNA Index System approval under the CODIS 20 standard.

Richard F. Selden founded the company and is its Chief Scientific Officer. He holds undergraduate and medical degrees from Harvard and was the first person to receive a PhD from Harvard’s Department of Genetics, and he is an inventor on a long list of US patents. Eugene Tan and Hua Jiang are both long term Ande scientists; all three are co-authors on the FlexPlex27 paper that describes the 27 locus assay used throughout the examples in this specification, which covers 23 autosomal loci, three Y-chromosomal loci and amelogenin and is designed to produce data compatible with the expanded CODIS core loci as well as the national databases of Australia, Canada, China, Germany, New Zealand and the United Kingdom.

The priority position is set out in the specification itself rather than having to be pieced together. Paragraph [0001] claims priority to US provisional applications 62/531,711 filed 12 July 2017 and 62/542,099 filed 7 August 2017. Paragraph [0003] states that AU 2018300157 is the Australian national phase entry of PCT/US2018/041838, published as WO 2019/014455. Paragraph [0002] records that AU 2024203902 was divided out of AU 2018300157, and that the present application has in turn been divided out of AU 2024203902. That makes the priority country the United States.

Why It Matters

The commercial argument for rapid DNA is that the profile is generated where the decision is made, not three weeks later in a laboratory queue. The specification points to the United States Department of Defense, FBI and Department of Homeland Security collaborating on requirements for such systems, the FBI’s creation of the Rapid DNA Index System, and the unanimous passage of the Rapid DNA Act of 2017 as evidence that profile generation outside the laboratory is becoming routine.

That premise puts pressure on exactly the step this claim addresses. In a laboratory, a flat-topped peak is an annoyance: quantify the extract again, dilute it, re-amplify, re-inject. At a disaster victim identification site or a crime scene there may be one swab, one bone fragment, one run, and no analyst on hand to decide what to do about it. The specification’s own framing is that the long-felt need is for a system that “effectively expands the dynamic range” of the process, and makes the quantitation step optional. Recovering a usable number from a clipped peak is a small piece of that, but it is the piece that decides whether a locus is called or written off.

It is worth being clear about what is and is not contentious here. Automating the interpretive judgement of a forensic analyst is genuinely disputed territory, which is why the specification has to quote the FBI saying that no expert system is approved for casework. Claim 1 sits on the safer side of that line. It does not decide whether a peak is an allele, a stutter artefact or a pull-up. It reconstructs a measurement that the hardware failed to record, using a ratio drawn from the same peak, and hands the result to whatever calls the alleles. Whether a reconstructed peak height is admissible evidence is a question the specification does not address, and it is the question a defence expert would ask first.

Related Concepts

• Short tandem repeat analysis – the identification method whose electropherogram peaks this system is correcting.
• Electropherogram – the trace of fluorescence peaks that the saturated detector distorts.
• Dynamic range – the span between the smallest and largest signal a detector can record, which is the limit being worked around.
• Clipping – the general signal processing name for the flat-topping that happens when a signal exceeds what the system can represent.
• Fluorophore – the dye attached to the PCR primers whose emission spills across more than one detector, which is what makes the cross detector ratio possible.
• Rapid DNA – the field-deployed analysis category the applicant’s instruments belong to.

---

AU 2026202193 was published in the Australian Official Journal of Patents on 9 April 2026 and is open for public inspection. Patent applications represent inventions that are sought to be protected and do not necessarily reflect commercially available products.