Futurists have anticipated that novel autonomous technologies, embedded with machine learning (ML), will substantially influence healthcare. ML is focused on making predictions as accurate as possible, while traditional statistical models are aimed at inferring relationships between variables. The benefits of ML comprise flexibility and scalability compared with conventional statistical approaches, which makes it deployable for several tasks, such as diagnosis and classification, and survival predictions. However, much of ML-based analysis remains scattered, lacking a cohesive structure. There is a need to evaluate and compare the performance of well-developed conventional statistical methods and ML on patient outcomes, such as survival, response to treatment, and patient-reported outcomes (PROs). In this article, we compare the usefulness and limitations of traditional statistical methods and ML, when applied to the medical field. Traditional statistical methods seem to be more useful when the number of cases largely exceeds the number of variables under study and a priori knowledge on the topic under study is substantial such as in public health. ML could be more suited in highly innovative fields with a huge bulk of data, such as omics, radiodiagnostics, drug development, and personalized treatment. Integration of the two approaches should be preferred over a unidirectional choice of either approach.

Autism diagnosis is moving from the identification of common inherited genetic variants to a systems biology approach. The aims of the study were to explore metabolic perturbations in autism, to investigate whether the severity of autism core symptoms may be associated with specific metabolic signatures; and to examine whether the urine metabolome discriminates severe from mild-to-moderate restricted, repetitive, and stereotyped behaviors. We enrolled 57 children aged 2–11 years; thirty-one with idiopathic autism and twenty-six neurotypical (NT), matched for age and ethnicity. The urine metabolome was investigated by gas chromatography-mass spectrometry (GC-MS). The urinary metabolome of autistic children was largely distinguishable from that of NT children; food selectivity induced further significant metabolic differences. Severe autism spectrum disorder core deficits were marked by high levels of metabolites resulting from diet, gut dysbiosis, oxidative stress, tryptophan metabolism, mitochondrial dysfunction. The hierarchical clustering algorithm generated two metabolic clusters in autistic children: 85–90% of children with mild-to-moderate abnormal behaviors fell in cluster II. Our results open up new perspectives for the more general understanding of the correlation between the clinical phenotype of autistic children and their urine metabolome. Adipic acid, palmitic acid, and 3-(3-hydroxyphenyl)-3-hydroxypropanoic acid can be proposed as candidate biomarkers of autism severity.

Metabolomics is an expanding discipline in biology. It is the process of portraying the phenotype of a cell, tissue or species organism using a comprehensive set of metabolites. Therefore, it is of interest to understand complex systems such as metabolomics using a scale-free topology. Genetic networks and the World Wide Web (WWW) are described as networks with complex topology. Several large networks have vertex connectivity that goes beyond a scale-free power-law distribution. It is observed that (a) networks expand constantly by the addition of recent vertices, and (b) recent vertices attach preferentially to sites that are already well connected. Scalefree networks are determined with precision using vital features such as a structure, a disease and a patient. This is pertinent to the understanding of complex systems such as metabolomics. Hence, we describe the relevance of scale-free networks in the understanding of metabolomics in this article.

Amyloid PET imaging has been crucial for detecting the accumulation of amyloid beta (Aβ) deposits in the brain and to study Alzheimer’s disease (AD). We performed a genome-wide association study on the largest collection of amyloid imaging data (N = 13,409) to date, across multiple ethnicities from multicenter cohorts to identify variants associated with brain amyloidosis and AD risk. We found a strong APOE signal on chr19q.13.32 (top SNP: APOE ɛ4; rs429358; β = 0.35, SE = 0.01, P = 6.2 × 10–311, MAF = 0.19), driven by APOE ɛ4, and five additional novel associations (APOE ε2/rs7412; rs73052335/rs5117, rs1081105, rs438811, and rs4420638) independent of APOE ɛ4. APOE ɛ4 and ε2 showed race specific effect with stronger association in Non-Hispanic Whites, with the lowest association in Asians. Besides the APOE, we also identified three other genome-wide loci: ABCA7 (rs12151021/chr19p.13.3; β = 0.07, SE = 0.01, P = 9.2 × 10–09, MAF = 0.32), CR1 (rs6656401/chr1q.32.2; β = 0.1, SE = 0.02, P = 2.4 × 10–10, MAF = 0.18) and FERMT2 locus (rs117834516/chr14q.22.1; β = 0.16, SE = 0.03, P = 1.1 × 10–09, MAF = 0.06) that all colocalized with AD risk. Sex-stratified analyses identified two novel female-specific signals on chr5p.14.1 (rs529007143, β = 0.79, SE = 0.14, P = 1.4 × 10–08, MAF = 0.006, sex-interaction P = 9.8 × 10–07) and chr11p.15.2 (rs192346166, β = 0.94, SE = 0.17, P = 3.7 × 10–08, MAF = 0.004, sex-interaction P = 1.3 × 10–03). We also demonstrated that the overall genetic architecture of brain amyloidosis overlaps with that of AD, Frontotemporal Dementia, stroke, and brain structure-related complex human traits. Overall, our results have important implications when estimating the individual risk to a population level, as race and sex will needed to be taken into account. This may affect participant selection for future clinical trials and therapies.