Back to Bigger Picture

BIGGER PICTURE Earlier diagnosis

From a genome’s thousands of possibilities to a few genes to inspect

For a critically ill child, sequencing DNA can be faster than interpreting what it means. GEM combines genetic variation with the child’s clinical features to bring plausible causes near the top of a review list.

A newborn’s small hand holds a gloved adult finger.Original photograph
Context photograph supplied by University of Utah Health with its GEM research release. It illustrates newborn care; it is not identified as a GEM study participant or a documented diagnosis.Charlie Ehlert / University of Utah HealthEditorial reuse with attribution
Previously diagnosed cases
179 in two cohorts
Cause near the top
Causal gene among top two in 92%
Review shortlist
Median of three candidate genes per case

How the idea works

  1. Combine DNA and symptoms

    Sequencing and clinical descriptions supply complementary clues.

  2. Compare with disease knowledge

    The model weighs variants, inheritance and phenotype together.

  3. Review a short list

    Specialists investigate the highest-ranked gene and disease candidates.

Reading the letters is not the same as understanding them

A genome contains many differences that do not explain a child’s illness. The diagnostic challenge is to connect a plausible variant, its inheritance and the observed symptoms. GEM integrates these clues with existing disease knowledge to nominate genes and disorders for expert review.

Sources: [1][2]

Clinical notes become part of the search

The study tested both manually prepared symptom descriptions and descriptions extracted from medical notes by clinical language processing. GEM also considered larger structural DNA changes, not just single-letter variants. This broadens the search while reducing the need to inspect every possible gene equally.

Sources: [2]

Ask whether the known answer rises to the top

The researchers benchmarked 119 previously diagnosed cases and validated the method in sixty more from five centres. Across these 179 cases, the causative gene appeared among the first two candidates in 92%. A median of three genes was prioritised for review, giving specialists a compact starting point.

Sources: [1][2]

Known diagnoses and new discoveries are different tests

The 92% figure is not a diagnosis rate for unselected sick children. When the team separately revisited fourteen unsolved cases, it reported one new finding. The achievement is stronger prioritisation with traceable clinical context; final interpretation and the link to treatment remain specialist work.

Sources: [2]

The original sources

  1. University of Utah HealthArtificial Intelligence-Based Technology Quickly Identifies Genetic Causes of Serious Disease
  2. Genome MedicineArtificial intelligence enables comprehensive genome interpretation and nomination of candidate diagnoses for rare genetic diseases

What to explore next

Sweden tested AI in breast screening—and found 29% more cancersA robotic boot learns from forty-four steps at a time
Explore all 99 applications