NOCIt relies on calibration curves to model key relationships: allele peak height vs. DNA template mass, stutter height vs. parent peak height, and baseline noise/dropout rates.
After the calibration curves are calculated it is essential to review them and determine if they meet expectations or if additional samples or adjudication is required prior to saving the calib.zip file.
1. Assess the Spread and Density of Points - review each locus under the Allele Peak Parameter
- Check the mass range: Your single-source calibration samples must span the entire operational range of your laboratory's amplification protocol (typically ranging from low-template stochastic levels like 0.016 ng up to optimal targets like 0.25 ng or higher). [1] Figure 1
- Identify data gaps: If the plot shows large blank gaps along the X-axis (DNA mass or peak height), the software has to rely on extrapolation. More samples are needed specifically targeting those sparse mass levels to stabilize the regression curves. Figure 2 is an example of a data gap.
2. Evaluate Residual Variance and Noise Modeling - review each locus under the Noise Peak Parameter
- Look at the baseline variance: NOCIt calculates variance in peak heights to establish signal-to-noise thresholds. If the residuals (the distance of your data points from the fitted curve) expand wildly or look erratic at certain heights, it indicates unstable variance. [1]
- Assess low-template clusters: If you do not have enough low-mass samples, the software cannot accurately model dropout probabilities or baseline noise. More samples are needed if the low-end curve looks poorly defined or doesn't reflect your actual analytical threshold. [1, 2] See figures 1 and 3
3. Review Stutter and Drop-out Regression Fits
- Stutter regression line: Ensure that forward and reverse stutter data points are numerous enough to form a clear, tight linear or curvilinear trend against parent allele heights. Outliers are normal, but a complete scatter with no visible trend line means the sample size is inadequate. [1, 2]
- Drop-out sigmoidal curve: The drop-out curve should smoothly transition from 0% probability at high template masses to 100% probability at ultra-low masses (See figure 4). If this curve transitions abruptly or fits poorly, more mid-to-low mass samples must be added to smooth out the probability distribution.
Figure 1. The plot has an even distribution of data points and contains data points across the range of masses, including low (Assess low-template clusters) necessary to model dropout.
Figure 2. The plot has a gap in the spread and low density at the upper mass, indicating that additional samples are required to develop the model at this locus.
Figure 3 Reflects the noise range, some outliers above typical analytical threshold, but evenly distributed and no outliers excessively higher than analytical thresholds.
Figure 4. Smooth sigmoidal curve to model drop-out.
1 NOCIt – a computational tool to infer the number of contributors to a forensic DNA sample Harish Swaminathan , Catherine M. Grgicak2, Muriel Medard , and Desmond S. Lun
2 https://go.softgenetics.com/guides/determining-number-of-contributors-in-dna-mixtures