Facilitating NGS-Based Screening of Genetic Disorders Using -AI-Driven Bioinformatics ^†

Abstract

Next-Generation Sequencing (NGS) is used as a diagnostic strategy for identifying pathogenic genetic variants in children and adults. However, the analysis is complex, requiring specialized bioinformaticians, and it can take weeks to finalize one study. This has been a limiting factor for the application of NGS in the screening of populations for rare genetic diseases. In this work, we show two case studies, where we applied an AI-driven bioinformatics framework in a diagnostic and a preventive scenario, respectively. The AI analysis was accurate and substantially faster than using conventional bioinformatics tools. Our results support the concept that AI-driven bioinformatics is a scalable solution for rendering accurate results and enabling a more widely available genetic screening for rare diseases.

1. Introduction

Whole Exome Sequencing (WES) using Next-Generation Sequencing (NGS) is a clinically accepted diagnostic technology for the identification of pathogenic genetic variants in children and adults [1]. Finding gene-function-disruptive variants (SNPs and INDELs) in sequences is fundamental in determining the cause of the genetic disease and for genetic counselling consultations. Additionally, the application of this method at the pre-conception stage can also enable parents to make informed decisions regarding the possible birth of children with a particular genetic disease. Databases such as ClinVar and OMIM have been accumulating information on an ever-increasing number of new pathogenic variants [2]. In these databases, gene–disease associations have also been growing over time, leading to more than eight thousand having been already reported [3]. Public and private healthcare facilities are beginning to use these data as a front-line tool over conventional techniques to diagnose pediatric rare genetic diseases [1,4]. However, the analysis of WES using bioinformatics is complex and requires specialist skills and training, hence it can take several weeks from sample to diagnosis [5]. The relative complexity associated with the high labor intensity is a substantial bottleneck in the field, leading to a heavy cost in human resources. This has been a limiting factor for the screening and prevention of rare diseases in the general population. Artificial Intelligence (AI) is considered to be a solution for automating complex analysis and decision-making [6]. In this work, we present two case studies where we applied an AI-driven bioinformatics framework in a diagnostic and a preventive scenario, respectively.

2. Methodology

2.1. Clinical Samples and Sequencing

Saliva samples were collected in DNA/RNA saliva collection tubes (GeneFix™, Isohelix) using the commercial ExoMart and SureMart kits from MolMart Ltd., Manchester, United Kingdom. Relevant clinical data were submitted by the referring clinician into the MolMart online form for kit activations (https://molmartgenomics.com, accessed on 27 October 2022). WES was performed by NGS using the Illumina platform. The exome library was prepared with Agilent’s SureSelect V6+UTR-post kit.

2.2. Bioinformatics

Variant Calling Files (VCF) were generated from FASTQ files using a standard bioinformatics pipeline [7,8]. BWA (Burrows–Wheeler Alignment Tool) software version 0.7.12 and reference human genome version hg38 were used for read mapping and alignment. Variant calling and variant annotation of genetic modifications was made using GATK (Genome Analysis Toolkit) software version 3.4.0 and SnpEff version 4.1, respectively. The MolMart Artificial Intelligence Analyst (MAIA) was used for pathogenic gene variant candidate identification and ranking on the clinical observations of the Variant Calling Files (VCF). Clinical observation matching and pathogenic scoring were performed by MAIA, considering both experimental evidence on databases and sequence predictions.

3. Results

We applied an AI-driven bioinformatics framework to analyze two case studies, one a diagnostic (Case Study 1) and the other a preventive scenario (Case Study 2).

3.1. Case Study 1

An 8-month-old infant was referred for genetic testing with hypotonia, delayed development, hepatosplenomegaly and strabismus. We applied AI-driven bioinformatics on the sequenced exome containing about 114,000 gene variants, taking into account the clinical phenotype (Figure 1). From all gene variants, the AI took ~5 s to identify a total of 757 putative pathogenic variants, where only 15 had high-scoring matches on disease database annotations that related to the clinical observations. Furthermore, the top-ranked variant (Figure 1) was the one chosen by independent molecular geneticists as causative of the phenotype by manually checking in the OMIM and ClinVar databases.

3.2. Case Study 2

In this case, we screened a healthy couple at the pre-conception stage for their potential risk of having a child affected by a genetic disease. We applied AI-driven bioinformatics on the male and female exomes containing about 112,000 and 113,000 gene variants, respectively (Figure 2). The AI took ~12 s to identify six putative pathogenic gene variants that can be transmitted from both males and females. From these, only one raised some concern based on strong gene–disease association evidence, with an estimated probability of 23% of having a child with mannose binding deficiency.

4. Conclusions

The case studies shown here demonstrate that AI-driven bioinformatics analysis is substantially faster than conventional bioinformatics tools and platforms. Furthermore, our results support the concept that AI-driven bioinformatics is an accurate and scalable solution which can make population-wide genetic screening for rare diseases possible.