Skip to main content
Premium Trial:

Request an Annual Quote

Baylor Genetics Prepares for a ‘Genome World’ With Explainable AI-Powered Variant Interpretation

By Illumina

During a presentation at the American College of Medical Genetics and Genomics (ACMG) annual clinical genetics meeting in March, Dr. Christine Eng, chief medical officer and chief quality officer at Baylor Genetics, described how the laboratory chose to utilize Illumina Connected Software for a comprehensive bioinformatics workflow. Baylor employs high-accuracy artificial intelligence, referred to as explainable artificial intelligence (XAI), to prioritize variants that are most likely to solve a case and to support its findings with transparent logic and links to relevant data and guidelines. During the presentation, titled “Preparing for a Genome World,” Eng described how XAI and informatics automation are enabling Baylor Genetics to hasten and scale up clinical whole-genome sequencing while testing becomes more accessible and evidence for its utility grows. Whole-genome sequencing, or WGS, has proven critical for acute-care patients and those whose conditions have eluded diagnosis, according to Eng.

Baylor Genetics launched clinical WGS testing in 2019, initially to support its activities as the core sequencing lab for the NIH Undiagnosed Diseases Network, or UDN. The lab now offers WGS as a routine service for patients in neonatal and pediatric intensive care units, for patients with congenital anomalies, developmental delays, and intellectual disabilities as recommended by the ACMG, and in certain other adult and pediatric settings. “WGS has a very high level of utility when you're evaluating complex phenotypes, or especially when a diagnosis is needed urgently and you don't have time to do your typical round of different genetic testing,” Eng said. 

Several advantages contribute to the utility of WGS, she added. It can detect a range of variant types; it eliminates the need for testing with multiple panels; the sequencing results can be re-analyzed as variants become better understood or when there is a new cause for testing; the diagnostic yield is higher and, with no capture step, the turnaround time is shorter as compared to whole-exome sequencing and NGS panels; and it can provide dual diagnoses when a patient has two conditions and presents with a blended phenotype. Despite this utility, access to WGS is constrained by the expertise required for testing and interpretation, limited coverage by payors, and the need for specialized technological infrastructure. 

However, Eng said, these constraints are rapidly falling away as expertise disseminates, technology improves, and sequencing costs fall. Additionally, payors are adopting progressive approaches to expanding WGS coverage, with Medicaid policies in 10 states now covering rapid WGS in acute care settings and some commercial payors covering WGS in outpatient settings. Eng said that these changes, along with the growing understanding of causative variants across the genome, are bringing about a “genome world” in which WGS will inform diagnosis and treatment in a wider variety of settings and may supplant other testing modalities. 

Considering that their previous workflow for whole exomes, which make up only 2 percent of the genome, required manual evaluation of 400 to 700 variants per exome, Eng and her team recognized that variant interpretation for WGS posed a “significant bottleneck and scalability challenge” that would be unsustainable in a genome world. 

To scale up testing and improve turnaround time, the lab validated and implemented an automated workflow with Illumina. Following sequencing, secondary analysis software makes variant calls, including complex structural variants, which are fed into tertiary analysis for interpretation. The XAI compares variants with a subject’s phenotypic information to gene-disease data from public and internal databases and shortlists the top potential causal variants. The Baylor Genetics team initially subjected these variants of interest to two layers of review by PhD-level genomic scientists and board-certified lab directors, but the strong performance of the auto-analysis by the XAI model enabled the lab to remove one layer of review, allowing for further scaling.

Critically for Baylor Genetics, the XAI reveals its logic for flagging and prioritizing candidate variants. “It's not a black box,” Eng said. “It can be looked at by the reviewers, and all of the supporting information in the literature or in various databases is there for the analysts to review.” The software produces an evidence graph, citing information from public databases like ClinVar, internal Baylor Genetics datasets, and ACMG guidelines. The XAI works across variant types — SNVs, indels, CNVs, mtDNA, STRs, and SV insertions — ensuring the workflow assesses all clinically relevant variations in the sample. 

The Baylor Genetics lab validated their standardized bioinformatics process with a retrospective cohort of 180 rare genetic disease cases that were previously evaluated and solved with whole-exome sequencing and manual interpretation, publishing their results in the June 2023 issue of Genetics in Medicine. In more than 93 percent of singleton cases, the top 10 suggestions generated by XAI included the causative variants previously identified by manual curation. When exome data from both parents was available, causative variants were included in the top recommendations 98 percent of the time, with 97 percent accuracy score overall.

In a prospective cohort of 334 new patients, the XAI-assisted process resolved 28.7 percent of cases, with an additional 12.6 percent having a possible diagnosis pending further evaluation. This diagnostic rate is comparable to that of Baylor Genetics’ manual evaluation process for rare genetic diseases (25 to 36.7 percent) and average rates among other commercial labs in the US (26 to 30 percent).

With the lab’s standardized operating procedure and optimized informatics stack, Baylor Genetics can provide rapid WGS test reports for acute-care patients in as few as five calendar days and standard reports in three weeks, according to Eng. She added that using Illumina’s wet lab to dry lab connected platforms, including Baylor’s automated variant interpretation workflows, allows for a “fully integrated ecosystem” that simplifies setup and support, requires minimal touchpoints, and readily scales.

While awareness of WGS and its utility is growing among clinicians and payors, more effort is needed to further improve access, according to Eng. She said Baylor Genetics works to raise awareness by publishing peer-reviewed papers on WGS, presenting their work at conferences around the country, and serving the UDN. She said she expects that recommended uses and payor coverage will broaden as labs continue to establish the utility of WGS. Additionally, she and colleagues will work to improve the diagnostic yield of WGS testing, perhaps by incorporating RNA-seq or other modalities. 

This sponsored content is provided by an advertiser and published in collaboration with the GW Custom Solutions Group, a division of GenomeWeb. The content was not produced by the editors or reporters of GenomeWeb, 360Dx, or Precision Oncology News, and does not represent the views of these publications or GenomeWeb's parent company, Crain Communications Inc.