André Altmann

MOTIVATION: In life sciences, interpretability of machine learning models is as important as their prediction accuracy. Linear models are probably the most frequently used methods for assessing feature relevance, despite their relative inflexibility. However, in the past years effective estimators of feature relevance have been derived for highly complex or non-parametric models such as support vector machines and RandomForest (RF) models. Recently, it has been observed that RF models are biased in such a way that categorical variables with a large number of categories are preferred. RESULTS: In this work, we introduce a heuristic for normalizing feature importance measures that can correct the feature importance bias. The method is based on repeated permutations of the outcome vector for estimating the distribution of measured importance for each variable in a non-informative setting. The P-value of the observed importance provides a corrected measure of feature importance. We apply our method to simulated data and demonstrate that (i) non-informative predictors do not receive significant P-values, (ii) informative variables can successfully be recovered among non-informative variables and (iii) P-values computed with permutation importance (PIMP) are very helpful for deciding the significance of variables, and therefore improve model interpretability. Furthermore, PIMP was used to correct RF-based importance measures for two real-world case studies. We propose an improved RF model that uses the significant variables with respect to the PIMP measure and show that its prediction accuracy is superior to that of other existing models. AVAILABILITY: R code for the method presented in this article is available at http://www.mpi-inf.mpg.de/ approximately altmann/download/PIMP.R CONTACT: altmann@mpi-inf.mpg.de, laura.tolosi@mpi-inf.mpg.de SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

OBJECTIVE: The APOE4 allele is the strongest genetic risk factor for sporadic Alzheimer disease (AD). Case-control studies suggest the APOE4 link to AD is stronger in women. We examined the APOE4-by-sex interaction in conversion risk (from healthy aging to mild cognitive impairment (MCI)/AD or from MCI to AD) and cerebrospinal fluid (CSF) biomarker levels. METHODS: Cox proportional hazards analysis was used to compute hazard ratios (HRs) for an APOE-by-sex interaction on conversion in controls (n = 5,496) and MCI patients (n = 2,588). The interaction was also tested in CSF biomarker levels of 980 subjects from the Alzheimer's Disease Neuroimaging Initiative. RESULTS: Among controls, male and female carriers were more likely to convert to MCI/AD, but the effect was stronger in women (HR = 1.81 for women; HR = 1.27 for men; interaction: p = 0.011). The interaction remained significant in a predefined subanalysis restricted to APOE3/3 and APOE3/4 genotypes. Among MCI patients, both male and female APOE4 carriers were more likely to convert to AD (HR = 2.16 for women; HR = 1.64 for men); the interaction was not significant (p = 0.14). In the subanalysis restricted to APOE3/3 and APOE3/4 genotypes, the interaction was significant (p = 0.02; HR = 2.17 for women; HR = 1.51 for men). The APOE4-by-sex interaction on biomarker levels was significant for MCI patients for total tau and the tau-to-Aβ ratio (p = 0.009 and p = 0.02, respectively; more AD-like in women). INTERPRETATION: APOE4 confers greater AD risk in women. Biomarker results suggest that increased APOE-related risk in women may be associated with tau pathology. These findings have important clinical implications and suggest novel research approaches into AD pathogenesis.

Abstract Medical artificial intelligence (AI) offers great potential for recognizing signs of health conditions in retinal images and expediting the diagnosis of eye diseases and systemic disorders 1 . However, the development of AI models requires substantial annotation and models are usually task-specific with limited generalizability to different clinical applications 2 . Here, we present RETFound, a foundation model for retinal images that learns generalizable representations from unlabelled retinal images and provides a basis for label-efficient model adaptation in several applications. Specifically, RETFound is trained on 1.6 million unlabelled retinal images by means of self-supervised learning and then adapted to disease detection tasks with explicit labels. We show that adapted RETFound consistently outperforms several comparison models in the diagnosis and prognosis of sight-threatening eye diseases, as well as incident prediction of complex systemic disorders such as heart failure and myocardial infarction with fewer labelled data. RETFound provides a generalizable solution to improve model performance and alleviate the annotation workload of experts to enable broad clinical AI applications from retinal imaging.

During rest, brain activity is synchronized between different regions widely distributed throughout the brain, forming functional networks. However, the molecular mechanisms supporting functional connectivity remain undefined. We show that functional brain networks defined with resting-state functional magnetic resonance imaging can be recapitulated by using measures of correlated gene expression in a post mortem brain tissue data set. The set of 136 genes we identify is significantly enriched for ion channels. Polymorphisms in this set of genes significantly affect resting-state functional connectivity in a large sample of healthy adolescents. Expression levels of these genes are also significantly associated with axonal connectivity in the mouse. The results provide convergent, multimodal evidence that resting-state functional networks correlate with the orchestrated activity of dozens of genes linked to ion channel activity and synaptic function.

This review summarizes the last decade of work by the ENIGMA (Enhancing NeuroImaging Genetics through Meta Analysis) Consortium, a global alliance of over 1400 scientists across 43 countries, studying the human brain in health and disease. Building on large-scale genetic studies that discovered the first robustly replicated genetic loci associated with brain metrics, ENIGMA has diversified into over 50 working groups (WGs), pooling worldwide data and expertise to answer fundamental questions in neuroscience, psychiatry, neurology, and genetics. Most ENIGMA WGs focus on specific psychiatric and neurological conditions, other WGs study normal variation due to sex and gender differences, or development and aging; still other WGs develop methodological pipelines and tools to facilitate harmonized analyses of "big data" (i.e., genetic and epigenetic data, multimodal MRI, and electroencephalography data). These international efforts have yielded the largest neuroimaging studies to date in schizophrenia, bipolar disorder, major depressive disorder, post-traumatic stress disorder, substance use disorders, obsessive-compulsive disorder, attention-deficit/hyperactivity disorder, autism spectrum disorders, epilepsy, and 22q11.2 deletion syndrome. More recent ENIGMA WGs have formed to study anxiety disorders, suicidal thoughts and behavior, sleep and insomnia, eating disorders, irritability, brain injury, antisocial personality and conduct disorder, and dissociative identity disorder. Here, we summarize the first decade of ENIGMA's activities and ongoing projects, and describe the successes and challenges encountered along the way. We highlight the advantages of collaborative large-scale coordinated data analyses for testing reproducibility and robustness of findings, offering the opportunity to identify brain systems involved in clinical syndromes across diverse samples and associated genetic, environmental, demographic, cognitive, and psychosocial factors.