Training samples for improved rare molecule detection

Why use a training sample?

For applications that require high specificity, such as rare molecule detection, it is important to include a training sample (TS) in each of your Countable experiments.

Training samples enable run-specific calibration using the platform’s machine-learning algorithm, improving the classification of true single-molecule signals from assay noise, thereby increasing confidence in low-count detection. This calibration enhances the specificity of single-molecule detection, particularly in low-abundance samples (<1,000 counts per 50 µL reaction), where assay noise becomes an issue and where single-molecule specificity is critical for many applications. When properly designed, a training sample improves both the accuracy and precision of your assay.

Designing a training sample?

When designing a TS for your multiplex assay, follow these guidelines:

Ensure all fluorescent probes, oligos, and other reaction components are added to the TS to account for intrinsic assay noise. For instance, if you intend to run a 4-plex assay on your test samples, the training sample should be run with the same assay with all the oligos present.
The TS should closely match your test sample(s) characteristics, including DNA size and input mass, to optimize the machine-learning algorithm and ensure accurate rare molecule detection.
Each TS should contain every target intended to be detected by the assay. For each target, there should be between 1,000 - 100,000 counts. For rare variant, we recommend using synthetic dsDNA spiked into background DNA of the same source as the test samples.

Examples of training samples

Assay type	Assay description	DNA input example	Training sample examples
Variant Allele Frequency (VAF) measurement	2-plex assay measuring variant allele abundance relative to wildtype allele in cfDNA	cfDNA, <25 ng	Option 1 - common variant allele: Fragmented gDNA (to ~100-200bp) from a human cell line with 50% Minor Allele Frequency (MAF) of the SNV, where both alleles are present loaded at 25 ng per reaction resulting in ~4000 counts per allele

Option 2 - rare variant allele: cell-free DNA from healthy donors, or fragmented gDNA (~100-200bp) spiked with synthetic dsDNA carrying the rare variant at ~10,000 counts | | Rare pathogen detection | 3-plex assay detecting rare pathogen DNA against high concentration host DNA background | ~3 µg host DNA containing the rare pathogen, fragmented to ~100 bp | Fragmented Lambda DNA (~100 bp) using the same fragmentation method as test samples at 3 µg spiked with ~10,000 counts of synthetic dsDNA carrying the target pathogen sequence |

Critical training sample considerations

For each Countable analysis, a unique TS should be used and applied to every scan.
When working with multiplexed samples, you have the flexibility to select a training sample for one to four detection channels. This means you can choose a single training sample to apply across multiple channels, or, depending on your experimental design, select a distinct training sample for each of the four detection channels.
If your experiment involves multiple assays, we recommend creating a separate Sample Workbook for each assay. This allows you to specify a TS unique to that particular assay, ensuring the correct TS is applied to all samples within its respective analysis.

Indicate a training sample in the Countable Control Software

Open the Countable Control Software on the Countable System. Click on the ‘Build Sample Workbook’ **tile on the Countable Control Software Main Menu. In section 4, select ‘Use Training Samples (Optional)’ if you are including a TS to improve rare molecule detection. Refer to IFU003 Countable System for detailed instructions on how to set up a Countable Sample Workbook and image your samples.

Screenshot 2025-06-24 at 4.16.48 PM.png

What if a training sample is not included in a Countable experiment?

For samples with <1000 counts per 50 µL, if no training samples are indicated in the experiment or if the indicated training sample has less than 1000 counts, the Countable Control Software will report an ‘Observation’ flag in the Countable Analysis Summary file. Counts will still be reported. However, they should be interpreted with care, especially for rare molecule detection applications.