Machine Learning for Causal Inference in Biological Networks: Perspectives of This Challenge

Causal machine learning models for predicting low birth weight in midwife-led continuity care intervention in North Shoa Zone, Ethiopia

BACKGROUND

Low birth weight (LBW) is a critical global health issue that affects infants disproportionately, particularly in developing countries. This study adopted causal machine learning (CML) algorithms for predicting LBW in newborns, drawing from midwife-led continuity care (MLCC).

METHODS

A quasi-experimental study was carried out in the North Shoa Zone of Ethiopia from August 2019 to September 2020. A total of 1166 women were allocated into two groups. The first group, the MLCC group, received all their antenatal, labor, birth, and immediate post-natal care from a single midwife. The second group received care from various staff members at different times throughout their pregnancy and childbirth. In this study, CML was implemented to predict LBW. Data preprocessing, including data cleaning, was conducted. CML was then employed to identify the most suitable classifier for predicting LBW. Gradient boosting algorithms were used to estimate the causal effect of MLCC on LBW. Moreover, meta-learner algorithms were utilized to estimate the individual treatment effect (ITE), the average treatment effect (ATE), and performance. Moreover, meta-learner algorithms were utilized to estimate the individual treatment effect (ITE), the average treatment effect (ATE), and performance.

RESULTS

The study results revealed that Causal K-Nearest Neighbors (CKNN) was the most effective classifier based on accuracy and estimated LBW using a 94.52% accuracy, 90.25% precision, 92.57% recall, and an F1 score of 88.2%. Meconium aspiration, perinatal mortality, pregnancy-induced hypertension, vacuum babies in need of resuscitation, and previous surgeries on their reproductive organs were identified as the top five features affecting LBW. The estimated impact of MLCC versus other professional groups on LBW was analyzed using gradient boosting algorithms and was found to be 0.237. The estimated ATE for the S-learner was 0.284, which is lower than the true ATE of 0.216. Additionally, the estimated ITE for both the T-learner and X-learner was less than -0.5, indicating that mothers would not choose to participate in the MLCC program.

CONCLUSIONS

Based on these findings, the CKNN classifier demonstrated a higher accuracy and effectiveness. The S-learner and R-learner models, utilizing the XGBoost Regressor and BaseSRegressor, provided accurate estimations of ITE for assessing the impact of the MLCC program. Promoting the MLCC program could help stabilize LBW outcomes.

Artificial Intelligence (AI)-driven approach to climate action and sustainable development

Countries have pledged commitment to the 2030 Sustainable Development Goal (SDGs) and the Paris Agreement to combat climate change. To maximize synergies between SDGs and climate actions (CAs), we evaluate the alignment of national commitment to SDGs and emissions reduction targets by comparing action plans embodied in Voluntary National Review (VNR) reports and the Nationally Determined Contributions (NDCs) across 67 countries. An Artificial Intelligence (AI)-based approach is proposed in this study to explore the interconnectedness by applying machine learning classifier and natural language processing. Middle- and low-income countries with high emissions tend to have low NDC targets and contain similar information in VNR reports. High-income countries show less alignment between their NDCs and VNRs. The economic status of countries is found to be connected to their climate actions and SDGs alignment. Here, we demonstrate utility and promise in using AI techniques to unravel interactions between CA and SDG.

Unleashing the future: The revolutionary role of machine learning and artificial intelligence in drug discovery

Drug discovery is a complex and multifaceted process aimed at identifying new therapeutic compounds with the potential to treat various diseases. Traditional methods of drug discovery are often time-consuming, expensive, and characterized by low success rates. Because of this, there is an urgent need to improve the drug development process using new technologies. The integration of the current state-of-art of artificial intelligence (AI) and machine learning (ML) approaches with conventional methods will enhance the efficiency and effectiveness of pharmaceutical research. This review highlights the transformative impact of AI and ML in drug discovery, discussing current applications, challenges, and future directions in harnessing these technologies to accelerate the development of innovative therapeutics. We have discussed the latest developments in AI and ML technologies to streamline several stages of drug discovery, from target identification and validation to lead optimization and preclinical studies.

Identification of diagnostic candidate genes in COVID-19 patients with sepsis

PURPOSE

Coronavirus Disease 2019 (COVID-19) and sepsis are closely related. This study aims to identify pivotal diagnostic candidate genes in COVID-19 patients with sepsis.

PATIENTS AND METHODS

We obtained a COVID-19 data set and a sepsis data set from the Gene Expression Omnibus (GEO) database. Identification of differentially expressed genes (DEGs) and module genes using the Linear Models for Microarray Data (LIMMA) and weighted gene co-expression network analysis (WGCNA), functional enrichment analysis, protein-protein interaction (PPI) network construction, and machine learning algorithms (least absolute shrinkage and selection operator (LASSO) regression and Random Forest (RF)) were used to identify candidate hub genes for the diagnosis of COVID-19 patients with sepsis. Receiver operating characteristic (ROC) curves were developed to assess the diagnostic value. Finally, the data set GSE28750 was used to verify the core genes and analyze the immune infiltration.

RESULTS

The COVID-19 data set contained 3,438 DEGs， and 595 common genes were screened in sepsis. sepsis DEGs were mainly enriched in immune regulation. The intersection of DEGs for COVID-19 and core genes for sepsis was 329, which were also mainly enriched in the immune system. After developing the PPI network, 17 node genes were filtered and thirteen candidate hub genes were selected for diagnostic value evaluation using machine learning. All thirteen candidate hub genes have diagnostic value, and 8 genes with an Area Under the Curve (AUC) greater than 0.9 were selected as diagnostic genes.

CONCLUSION

Five core genes (CD3D, IL2RB, KLRC, CD5, and HLA-DQA1) associated with immune infiltration were identified to evaluate their diagnostic utility COVID-19 patients with sepsis. This finding contributes to the identification of potential peripheral blood diagnostic candidate genes for COVID-19 patients with sepsis.

Gene regulatory networks in disease and ageing

The precise control of gene expression is required for the maintenance of cellular homeostasis and proper cellular function, and the declining control of gene expression with age is considered a major contributor to age-associated changes in cellular physiology and disease. The coordination of gene expression can be represented through models of the molecular interactions that govern gene expression levels, so-called gene regulatory networks. Gene regulatory networks can represent interactions that occur through signal transduction, those that involve regulatory transcription factors, or statistical models of gene-gene relationships based on the premise that certain sets of genes tend to be coexpressed across a range of conditions and cell types. Advances in experimental and computational technologies have enabled the inference of these networks on an unprecedented scale and at unprecedented precision. Here, we delineate different types of gene regulatory networks and their cell-biological interpretation. We describe methods for inferring such networks from large-scale, multi-omics datasets and present applications that have aided our understanding of cellular ageing and disease mechanisms.

A molecular index for biological age identified from the metabolome and senescence-associated secretome in humans

Unlike chronological age, biological age is a strong indicator of health of an individual. However, the molecular fingerprint associated with biological age is ill-defined. To define a high-resolution signature of biological age, we analyzed metabolome, circulating senescence-associated secretome (SASP)/inflammation markers and the interaction between them, from a cohort of healthy and rapid agers. The balance between two fatty acid oxidation mechanisms, β-oxidation and ω-oxidation, associated with the extent of functional aging. Furthermore, a panel of 25 metabolites, Healthy Aging Metabolic (HAM) index, predicted healthy agers regardless of gender and race. HAM index was also validated in an independent cohort. Causal inference with machine learning implied three metabolites, β-cryptoxanthin, prolylhydroxyproline, and eicosenoylcarnitine as putative drivers of biological aging. Multiple SASP markers were also elevated in rapid agers. Together, our findings reveal that a network of metabolic pathways underlie biological aging, and the HAM index could serve as a predictor of phenotypic aging in humans.

The use of machine learning on administrative and survey data to predict suicidal thoughts and behaviors: a systematic review

Background

Machine learning is a promising tool in the area of suicide prevention due to its ability to combine the effects of multiple risk factors and complex interactions. The power of machine learning has led to an influx of studies on suicide prediction, as well as a few recent reviews. Our study distinguished between data sources and reported the most important predictors of suicide outcomes identified in the literature.

Objective

Our study aimed to identify studies that applied machine learning techniques to administrative and survey data, summarize performance metrics reported in those studies, and enumerate the important risk factors of suicidal thoughts and behaviors identified.

Methods

A systematic literature search of PubMed, Medline, Embase, PsycINFO, Web of Science, Cumulative Index to Nursing and Allied Health Literature (CINAHL), and Allied and Complementary Medicine Database (AMED) to identify all studies that have used machine learning to predict suicidal thoughts and behaviors using administrative and survey data was performed. The search was conducted for articles published between January 1, 2019 and May 11, 2022. In addition, all articles identified in three recently published systematic reviews (the last of which included studies up until January 1, 2019) were retained if they met our inclusion criteria. The predictive power of machine learning methods in predicting suicidal thoughts and behaviors was explored using box plots to summarize the distribution of the area under the receiver operating characteristic curve (AUC) values by machine learning method and suicide outcome (i.e., suicidal thoughts, suicide attempt, and death by suicide). Mean AUCs with 95% confidence intervals (CIs) were computed for each suicide outcome by study design, data source, total sample size, sample size of cases, and machine learning methods employed. The most important risk factors were listed.

Results

The search strategy identified 2,200 unique records, of which 104 articles met the inclusion criteria. Machine learning algorithms achieved good prediction of suicidal thoughts and behaviors (i.e., an AUC between 0.80 and 0.89); however, their predictive power appears to differ across suicide outcomes. The boosting algorithms achieved good prediction of suicidal thoughts, death by suicide, and all suicide outcomes combined, while neural network algorithms achieved good prediction of suicide attempts. The risk factors for suicidal thoughts and behaviors differed depending on the data source and the population under study.

Conclusion

The predictive utility of machine learning for suicidal thoughts and behaviors largely depends on the approach used. The findings of the current review should prove helpful in preparing future machine learning models using administrative and survey data.

Systematic review registration

https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42022333454 identifier CRD42022333454.

Machine-learning algorithms in screening for type 2 diabetes mellitus: Data from Fasa Adults Cohort Study

INTRODUCTION

The application of machine learning (ML) is increasingly growing in biomedical sciences. This study aimed to evaluate factors associated with type 2 diabetes mellitus (T2DM) and compare the performance of ML methods in identifying individuals with the disease in an Iranian setting.

METHODS

Using the baseline data from Fasa Adult Cohort Study (FACS) and in a sex-stratified manner, we studied factors associated with T2DM by applying seven different ML methods including Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbours (KNN), Gradient Boosting Machine (GBM), Extreme Gradient Boosting (XGB) and Bagging classifier (BAG). We further compared the performance of these methods; for each algorithm, accuracy, precision, sensitivity, specificity, F1 score, and Area Under Curve (AUC) were calculated.

RESULTS

10,112 participants were recruited between 2014 and 2016, of whom 1246 had T2DM at baseline. 4566 (45%) participants were males, aged between 35 and 70 years. For males, age, sugar consumption, and history of hospitalization were the most weighted variables regarding their importance in screening for T2DM using the GBM model, respectively; these variables were sugar consumption, urine blood, and age for females. GBM outperformed other models for both males and females with AUC of 0.75 (0.69-0.82) and 0.76 (0.71-0.80), and F1 score of 0.33 (0.27-0.39) and 0.42 (0.38-0.46), respectively. GBM also showed a sensitivity of 0.24 (0.19-0.29) and a specificity of 0.98 (0.96-1.0) in males and a sensitivity of 0.38 (0.34-0.42) and specificity of 0.92 (0.89-0.95) in females. Notably, close performance characteristics were detected among other ML models.

CONCLUSIONS

GBM model might achieve better performance in screening for T2DM in a south Iranian population.

Artificial intelligence for dementia research methods optimization

Artificial intelligence (AI) and machine learning (ML) approaches are increasingly being used in dementia research. However, several methodological challenges exist that may limit the insights we can obtain from high-dimensional data and our ability to translate these findings into improved patient outcomes. To improve reproducibility and replicability, researchers should make their well-documented code and modeling pipelines openly available. Data should also be shared where appropriate. To enhance the acceptability of models and AI-enabled systems to users, researchers should prioritize interpretable methods that provide insights into how decisions are generated. Models should be developed using multiple, diverse datasets to improve robustness, generalizability, and reduce potentially harmful bias. To improve clarity and reproducibility, researchers should adhere to reporting guidelines that are co-produced with multiple stakeholders. If these methodological challenges are overcome, AI and ML hold enormous promise for changing the landscape of dementia research and care. HIGHLIGHTS: Machine learning (ML) can improve diagnosis, prevention, and management of dementia. Inadequate reporting of ML procedures affects reproduction/replication of results. ML models built on unrepresentative datasets do not generalize to new datasets. Obligatory metrics for certain model structures and use cases have not been defined. Interpretability and trust in ML predictions are barriers to clinical translation.

Artificial intelligence for dementia prevention

INTRODUCTION

A wide range of modifiable risk factors for dementia have been identified. Considerable debate remains about these risk factors, possible interactions between them or with genetic risk, and causality, and how they can help in clinical trial recruitment and drug development. Artificial intelligence (AI) and machine learning (ML) may refine understanding.

METHODS

ML approaches are being developed in dementia prevention. We discuss exemplar uses and evaluate the current applications and limitations in the dementia prevention field.

RESULTS

Risk-profiling tools may help identify high-risk populations for clinical trials; however, their performance needs improvement. New risk-profiling and trial-recruitment tools underpinned by ML models may be effective in reducing costs and improving future trials. ML can inform drug-repurposing efforts and prioritization of disease-modifying therapeutics.

DISCUSSION

ML is not yet widely used but has considerable potential to enhance precision in dementia prevention.

HIGHLIGHTS

Artificial intelligence (AI) is not widely used in the dementia prevention field. Risk-profiling tools are not used in clinical practice. Causal insights are needed to understand risk factors over the lifespan. AI will help personalize risk-management tools for dementia prevention. AI could target specific patient groups that will benefit most for clinical trials.

What is circulating factor disease and how is it currently explained?

Nephrotic syndrome (NS) consists of the clinical triad of hypoalbuminaemia, high levels of proteinuria and oedema, and describes a heterogeneous group of disease processes with different underlying drivers. The existence of circulating factor disease (CFD) as a driver of NS has been epitomised by a subset of patients who exhibit disease recurrence after transplantation, alongside laboratory work. Several circulating factors have been proposed and studied, broadly grouped into protease components such as soluble urokinase-type plasminogen activator (suPAR), hemopexin (Hx) and calcium/calmodulin-serine protease kinase (CASK), and other circulating proteases, and immune components such as TNF-α, CD40 and cardiotrophin-like cytokine-1 (CLC-1). While currently there is no definitive way of assessing risk of CFD pre-transplantation, promising work is emerging through the study of 'multi-omic' bioinformatic data from large national cohorts and biobanks.

Use of big data and machine learning algorithms to extract possible treatment targets in neurodevelopmental disorders

Neurodevelopmental disorders (NDDs) impact multiple aspects of an individual's functioning, including social interactions, communication, and behaviors. The underlying biological mechanisms of NDDs are not yet fully understood, and pharmacological treatments have been limited in their effectiveness, in part due to the complex nature of these disorders and the heterogeneity of symptoms across individuals. Identifying genetic loci associated with NDDs can help in understanding biological mechanisms and potentially lead to the development of new treatments. However, the polygenic nature of these complex disorders has made identifying new treatment targets from genome-wide association studies (GWAS) challenging. Recent advances in the fields of big data and high-throughput tools have provided radically new insights into the underlying biological mechanism of NDDs. This paper reviews various big data approaches, including classical and more recent techniques like deep learning, which can identify potential treatment targets from GWAS and other omics data, with a particular emphasis on NDDs. We also emphasize the increasing importance of explainable and causal machine learning (ML) methods that can aid in identifying genes, molecular pathways, and more complex biological processes that may be future targets of intervention in these disorders. We conclude that these new developments in genetics and ML hold promise for advancing our understanding of NDDs and identifying novel treatment targets.

Recovery of dynamical similarity from lossy representations of collective behavior of midge swarms

Understanding emergent collective phenomena in biological systems is a complex challenge due to the high dimensionality of state variables and the inability to directly probe agent-based interaction rules. Therefore, if one wants to model a system for which the underpinnings of the collective process are unknown, common approaches such as using mathematical models to validate experimental data may be misguided. Even more so, if one lacks the ability to experimentally measure all the salient state variables that drive the collective phenomena, a modeling approach may not correctly capture the behavior. This problem motivates the need for model-free methods to characterize or classify observed behavior to glean biological insights for meaningful models. Furthermore, such methods must be robust to low dimensional or lossy data, which are often the only feasible measurements for large collectives. In this paper, we show that a model-free and unsupervised clustering of high dimensional swarming behavior in midges (Chironomus riparius), based on dynamical similarity, can be performed using only two-dimensional video data where the animals are not individually tracked. Moreover, the results of the classification are physically meaningful. This work demonstrates that low dimensional video data of collective motion experiments can be equivalently characterized, which has the potential for wide applications to data describing animal group motion acquired in both the laboratory and the field.

AI-powered therapeutic target discovery

Disease modeling and target identification are the most crucial initial steps in drug discovery, and influence the probability of success at every step of drug development. Traditional target identification is a time-consuming process that takes years to decades and usually starts in an academic setting. Given its advantages of analyzing large datasets and intricate biological networks, artificial intelligence (AI) is playing a growing role in modern drug target identification. We review recent advances in target discovery, focusing on breakthroughs in AI-driven therapeutic target exploration. We also discuss the importance of striking a balance between novelty and confidence in target selection. An increasing number of AI-identified targets are being validated through experiments and several AI-derived drugs are entering clinical trials; we highlight current limitations and potential pathways for moving forward.

A dynamical systems approach for multiscale synthesis of Alzheimer's pathogenesis

Alzheimer's disease (AD) is a spatially dynamic pathology that implicates a growing volume of multiscale data spanning genetic, cellular, tissue, and organ levels of the organization. These data and bioinformatics analyses provide clear evidence for the interactions within and between these levels. The resulting heterarchy precludes a linear neuron-centric approach and necessitates that the numerous interactions are measured in a way that predicts their impact on the emergent dynamics of the disease. This level of complexity confounds intuition, and we propose a new methodology that uses non-linear dynamical systems modeling to augment intuition and that links with a community-wide participatory platform to co-create and test system-level hypotheses and interventions. In addition to enabling the integration of multiscale knowledge, key benefits include a more rapid innovation cycle and a rational process for prioritization of data campaigns. We argue that such an approach is essential to support the discovery of multilevel-coordinated polypharmaceutical interventions.

Unified tumor growth mechanisms from multimodel inference and dataset integration

Mechanistic models of biological processes can explain observed phenomena and predict responses to a perturbation. A mathematical model is typically constructed using expert knowledge and informal reasoning to generate a mechanistic explanation for a given observation. Although this approach works well for simple systems with abundant data and well-established principles, quantitative biology is often faced with a dearth of both data and knowledge about a process, thus making it challenging to identify and validate all possible mechanistic hypothesis underlying a system behavior. To overcome these limitations, we introduce a Bayesian multimodel inference (Bayes-MMI) methodology, which quantifies how mechanistic hypotheses can explain a given experimental datasets, and concurrently, how each dataset informs a given model hypothesis, thus enabling hypothesis space exploration in the context of available data. We demonstrate this approach to probe standing questions about heterogeneity, lineage plasticity, and cell-cell interactions in tumor growth mechanisms of small cell lung cancer (SCLC). We integrate three datasets that each formulated different explanations for tumor growth mechanisms in SCLC, apply Bayes-MMI and find that the data supports model predictions for tumor evolution promoted by high lineage plasticity, rather than through expanding rare stem-like populations. In addition, the models predict that in the presence of cells associated with the SCLC-N or SCLC-A2 subtypes, the transition from the SCLC-A subtype to the SCLC-Y subtype through an intermediate is decelerated. Together, these predictions provide a testable hypothesis for observed juxtaposed results in SCLC growth and a mechanistic interpretation for tumor treatment resistance.

Drug Design and Success of Prospective Mouse In Vitro-In Vivo Extrapolation (IVIVE) for Predictions of Plasma Clearance (CLp) from Hepatocyte Intrinsic Clearance (CLint)

Hepatocyte intrinsic clearance (CL_int) and methods of in vitro-in vivo extrapolation (IVIVE) are often used to predict plasma clearance (CL_p) in drug discovery. While the prediction success of this approach is dependent on the chemotype, specific molecular properties and drug design features that govern these outcomes are poorly understood. To address this challenge, we investigated the success of prospective mouse CL_p IVIVE across 2142 chemically diverse compounds. Dilution scaling, which assumes that the free fraction in hepatocyte incubations (f_u,inc) is governed by binding to the 10% of serum in the incubation medium, was used as our default CL_p IVIVE approach. Results show that predictions of CL_p are better for smaller (molecular weight (MW) < 500 Da), less polar (total polar surface area (TPSA) < 100 Ų, hydrogen bond donor (HBD) ≤1, hydrogen bond acceptor (HBA) ≤ 6), lipophilic (log D > 3), and neutral compounds, with low HBD count playing the key role. If compounds are classified according to their chemical space, predictions were good for compounds resembling central nervous system (CNS) drugs [average absolute fold error (AAFE) of 2.05, average fold error (AFE) of 0.90], moderate for classical druglike compounds (according to Lipinski, Veber, and Ghose guidelines; AAFE of 2.55; AFE of 0.68), and poor for nonclassical "beyond the rule of 5" compounds (AAFE of 3.31; AFE of 0.41). From the perspective of measured druglike properties, predictions of CL_p were better for compounds with moderate-to-high hepatocyte CL_int (>10 μL/min/10⁶ cells), high passive cellular permeability (P_app > 100 nm/s), and moderate observed CL_p (5-50 mL/min/kg). Influences of plasma protein binding (f_u,p) and P-glycoprotein (Pgp) apical efflux ratio (AP-ER) were less pronounced. If the extended clearance classification system (ECCS) is applied, predictions were good for class 2 (P_app > 50 nm/s; neutral or basic; AAFE of 2.35; AFE of 0.70) and acceptable for class 1A compounds (AAFE of 2.98; AFE of 0.70). Classes 1B, 3 A/B, and 4 showed poor outcomes (AAFE > 3.80; AFE < 0.60). Functional groups trending toward weaker CL_p IVIVE were esters, carbamates, sulfonamides, carboxylic acids, ketones, primary and secondary amines, primary alcohols, oxetanes, and compounds liable to aldehyde oxidase metabolism, likely due to multifactorial reasons. Multivariate analysis showed that multiple properties are relevant, combining together to define the overall success of CL_p IVIVE. Our results indicate that the current practice of prospective CL_p IVIVE is suitable only for CNS-like compounds and well-behaved classical druglike space (e.g., high permeability or ECCS class 2) without challenging functional groups. Unfortunately, based on existing mouse data, prospective CL_p IVIVE for complex and nonclassical chemotypes is poor and hardly better than random guessing. This is likely due to complexities such as extrahepatic metabolism and transporter-mediated disposition which are poorly captured by this methodology. With small-molecule drug discovery increasingly evolving toward nonclassical and complex chemotypes, existing CL_p IVIVE methodology will require improvement. While empirical correction factors may bridge the gap in the near future, improved and new in vitro assays, data integration models, and machine learning (ML) methods are increasingly needed to address this challenge and reduce the number of nonclinical pharmacokinetic (PK) studies.

Tumor heterogeneity: preclinical models, emerging technologies, and future applications

Heterogeneity describes the differences among cancer cells within and between tumors. It refers to cancer cells describing variations in morphology, transcriptional profiles, metabolism, and metastatic potential. More recently, the field has included the characterization of the tumor immune microenvironment and the depiction of the dynamics underlying the cellular interactions promoting the tumor ecosystem evolution. Heterogeneity has been found in most tumors representing one of the most challenging behaviors in cancer ecosystems. As one of the critical factors impairing the long-term efficacy of solid tumor therapy, heterogeneity leads to tumor resistance, more aggressive metastasizing, and recurrence. We review the role of the main models and the emerging single-cell and spatial genomic technologies in our understanding of tumor heterogeneity, its contribution to lethal cancer outcomes, and the physiological challenges to consider in designing cancer therapies. We highlight how tumor cells dynamically evolve because of the interactions within the tumor immune microenvironment and how to leverage this to unleash immune recognition through immunotherapy. A multidisciplinary approach grounded in novel bioinformatic and computational tools will allow reaching the integrated, multilayered knowledge of tumor heterogeneity required to implement personalized, more efficient therapies urgently required for cancer patients.

An introduction to causal inference for pharmacometricians

As formal causal inference begins to play a greater role in disciplines that intersect with pharmacometrics, such as biostatistics, epidemiology, and artificial intelligence/machine learning, pharmacometricians may increasingly benefit from a basic fluency in foundational causal inference concepts. This tutorial seeks to orient pharmacometricians to three such fundamental concepts: potential outcomes, g-formula, and directed acyclic graphs (DAGs).