Lingo: an automated, web-based deep phenotyping platform for language ability

Language abilities are highly heritable, yet their genetic architecture remains poorly understood due to limitations in scalable phenotyping. Traditional clinical assessments are time-intensive (often 1-3 hours) and require specialized personnel, fundamentally constraining the sample sizes necessary for robust genetic discovery. Here we introduce Lingo, an open-source web-based platform that captures comprehensive language and cognitive phenotypes in approximately 30 minutes through seven tasks administered remotely. Critically, comparative power analyses demonstrate that Lingo-derived phenotypes achieve nearly 2-fold greater statistical power than clinical questionnaire measures and >10-fold improvement over self-reported language diagnoses for detecting polygenic score associations. This translates to meaningful sample size reductions: 1,000 participants using Lingo is equivalent to 1,730 with in-depth questionnaires or > 13,000 people with self-reported language impairment. Analyzing Lingo data from > 2,000 adults, we identify four interpretable factors with high test-retest reliability (r = 0.69-0.79): narrative fluency, reading fluency, phonemic fluency, and general cognitive ability (g). The g factor demonstrates strong concurrent validity with clinical Wechsler full-scale IQ assessment (r = 0.78). These factors reveal distinct psychiatric and genetic profiles: g associates with externalizing behaviors and ADHD polygenic propensity, while phonemic fluency links specifically to withdrawn behavior as well as depression and schizophrenia polygenic scores. Rare variant burden analysis of Lingo scores identifies novel candidate genes (NGB and GLS) and implicates ATP metabolism and white matter pathways in language ability. These results establish Lingo as a transformative tool that makes previously impractical genetic studies feasible by enabling adequately powered studies with dramatically reduced sample sizes and costs.