Machine learning algorithms reveal the secrets of mitochondrial dynamics.

Mitochondria exist as dynamic networks whose morphology is driven by the complex interplay between fission and fusion events. Failure to modulate these processes can be detrimental to human health as evidenced by dominantly inherited, pathogenic variants in OPA1, an effector enzyme of mitochondrial fusion, that lead to network fragmentation, cristae dysmorphology and impaired oxidative respiration, manifesting typically as isolated optic atrophy. However, a significant number of patients develop more severe, systemic phenotypes, although no genetic modifiers of OPA1-related disease have been identified to date. In this issue of EMBO Molecular Medicine, supervised machine learning algorithms underlie a novel tool that enables automated, high throughput and unbiased screening of changes in mitochondrial morphology measured using confocal microscopy. By coupling this approach with a bespoke siRNA library targeting the entire mitochondrial proteome, the work described by Cretin and colleagues yielded significant insight into mitochondrial biology, discovering 91 candidate genes whose endogenous depletion can remedy impaired mitochondrial dynamics caused by OPA1 deficiency.