Inborn errors of metabolism (IEM) are inherited conditions caused by genetic defects in enzymes or cofactors. These defects result in a specific metabolic fingerprint in patient body fluids, showing accumulation of substrate or lack of an end-product of the defective enzymatic step. Untargeted metabolomics has evolved as a high throughput methodology offering a comprehensive readout of this metabolic fingerprint. This makes it a promising tool for diagnostic screening of IEM patients. However, the size and complexity of metabolomics data have posed a challenge in translating this avalanche of information into knowledge, particularly for clinical application. We have previously established next-generation metabolic screening (NGMS) as a metabolomics-based diagnostic tool for analyzing plasma of individual IEM-suspected patients. To fully exploit the clinical potential of NGMS, we present a computational pipeline to streamline the analysis of untargeted metabolomics data. This pipeline allows for time-efficient and reproducible data analysis, compatible with ISO:15189 accredited clinical diagnostics. The pipeline implements a combination of tools embedded in a workflow environment for large-scale clinical metabolomics data analysis. The accompanying graphical user interface aids end-users from a diagnostic laboratory for efficient data interpretation and reporting. We also demonstrate the application of this pipeline with a case study and discuss future prospects.

Allele calling AI models have the potential to improve the efficiency and accuracy of forensic DNA profile analysis. In order to train and validate such AI models, large, high-quality and representative labelled datasets of DNA profiles are required. Manually producing such datasets is resource intensive. We present here, DNAnnotator, a python-based labelling tool for DNA profiles to assist any forensic DNA laboratory with creating their own annotated datasets, for training and validating AI models for allele calling. We also present our experience using DNAnnotator to create a labelled dataset of 1,776 profiles consisting of reference and trace profiles from routine forensic casework and in-house research data. Annotating all peaks in a profile took on average 30 minutes using DNAnnotator. Pre-labelling the profile with a trained model reduced annotation time by 40%. Overall, label type (allele, stutter and various artefacts) agreement was found to be 97% across three annotators annotating profiles in common. Disagreement was slightly greater in trace profiles when compared to reference profiles and increased as electropherogram (EPG) peaks decreased in height. Label disagreement was principally due to stutter peaks, reflecting the known complexity of interpretation by human DNA profile analysts. These findings highlight the potential of AI models to support more efficient and consistent EPG peak interpretation. DNAnnotator facilitates the creation of labelled training datasets, supports the training and validation of AI models for DNA profile analysis, and is openly available on GitHub for use by forensic DNA laboratories.