ABSTRACT
Semi-supervised learning (SSL) aims to train a machine learning (ML) model using both labeled and unlabeled data. While the unlabeled data have been used in various ways to improve the prediction accuracy, the reason why unlabeled data could help is not fully understood. One interesting and promising direction is to understand SSL from a causal perspective. In light of the independent causal mechanisms (ICM) principle, the unlabeled data can be helpful when the label causes the features but not vice versa. However, the causal relations between the features and labels can be complex in real world applications. In this article, we propose an SSL framework that works with general causal models in which the variables have flexible causal relations. More specifically, we explore the causal graph structures and design corresponding causal generative models which can be learned with the help of unlabeled data. The learned causal generative model can generate synthetic labeled data for training a more accurate predictive model. We verify the effectiveness of our proposed method by empirical studies on both simulated and real data.
ABSTRACT
BACKGROUND: Attention-deficit/hyperactivity disorder (ADHD) and autism spectrum disorder (ASD) are neurodevelopmental conditions with early life origins. Alterations in blood lipids have been linked to ADHD and ASD; however, prospective early life data are limited. This study examined (i) associations between the cord blood lipidome and ADHD/ASD symptoms at 2 years of age, (ii) associations between prenatal and perinatal predictors of ADHD/ASD symptoms and cord blood lipidome, and (iii) mediation by the cord blood lipidome. METHODS: From the Barwon Infant Study cohort (1074 mother-child pairs, 52.3% male children), child circulating lipid levels at birth were analysed using ultra-high-performance liquid chromatography-tandem mass spectrometry. These were clustered into lipid network modules via Weighted Gene Correlation Network Analysis. Associations between lipid modules and ADHD/ASD symptoms at 2 years, assessed with the Child Behavior Checklist, were explored via linear regression analyses. Mediation analysis identified indirect effects of prenatal and perinatal risk factors on ADHD/ASD symptoms through lipid modules. FINDINGS: The acylcarnitine lipid module is associated with both ADHD and ASD symptoms at 2 years of age. Risk factors of these outcomes such as low income, Apgar score, and maternal inflammation were partly mediated by higher birth acylcarnitine levels. Other cord blood lipid profiles were also associated with ADHD and ASD symptoms. INTERPRETATION: This study highlights that elevated cord blood birth acylcarnitine levels, either directly or as a possible marker of disrupted cell energy metabolism, are on the causal pathway of prenatal and perinatal risk factors for ADHD and ASD symptoms in early life. FUNDING: The foundational work and infrastructure for the BIS was sponsored by the Murdoch Children's Research Institute, Deakin University, and Barwon Health. Subsequent funding was secured from the Minderoo Foundation, the European Union's Horizon 2020 research and innovation programme (ENDpoiNTs: No 825759), National Health and Medical Research Council of Australia (NHMRC) and Agency for Science, Technology and Research Singapore [APP1149047], The William and Vera Ellen Houston Memorial Trust Fund (via HOMER Hack), The Shepherd Foundation, The Jack Brockhoff Foundation, the Scobie & Claire McKinnon Trust, the Shane O'Brien Memorial Asthma Foundation, the Our Women Our Children's Fund Raising Committee Barwon Health, the Rotary Club of Geelong, the Ilhan Food Allergy Foundation, Geelong Medical and Hospital Benefits Association, Vanguard Investments Australia Ltd, the Percy Baxter Charitable Trust, and Perpetual Trustees.
