Prediction of clinically relevant drug‐induced liver injury from structure using machine learning

Abstract

Drug‐induced liver injury (DILI) is the most common cause of acute liver failure and often responsible for drug withdrawals from the market. Clinical manifestations vary, and toxicity may or may not appear dose‐dependent. We present several machine‐learning models (decision tree induction, k‐nearest neighbor, support vector machines, artificial neural networks) for the prediction of clinically relevant DILI based solely on drug structure, with data taken from published DILI cases. Our models achieved corrected classification rates of up to 89%. We also studied the association of a drug's interaction with carriers, enzymes and transporters, and the relationship of defined daily doses with hepatotoxicity. The results presented here are useful as a screening tool both in a clinical setting in the assessment of DILI as well as in the early stages of drug development to rule out potentially hepatotoxic candidates.

In vitro hepatotoxicity of Petasites hybridus extract (Ze 339) depends on the concentration, the cytochrome activity of the cell system, and the species used

“…Besides daily dose (≥100 mg) and lipophilicity (LogP≥3), bioactivation was identified as a significant risk factor for the development of drug‐induced hepatotoxicity (Chen, Borlak, & Tong, ; Yu et al, ). The importance of these risk factors was also recently confirmed by us in larger dataset (Hammann, Schöning, & Drewe, ). This may explain the observed species‐specific effects in preclinical in vivo studies, with rats being the most sensitive species, followed by dogs, and the good tolerability profile in clinical studies and post‐marketing observations.…”

Artificial Intelligence in Drug Discovery: A Comprehensive Review of Data-driven and Machine Learning Approaches

“…DILI is one of the main reasons for the withdrawal of marketed drugs. Therefore, there have been many in silico research studies to make precise predictive models and find the patterns of hepatotoxic compounds [159][160][161][162][163][164][165]. Kotsampasakou et al [161] highlighted the importance of data curation by curating 1,547 compounds from various sources and testing the performance of machine learning-based prediction models.…”

“…The author collected data from various sources; LTKB-BD, DrugBank, and literatures to construct and validate the model. Hammann et al [163] constructed a DILI prediction model based on DILI annotated drugs by using physicochemical descriptors and machine learning methods. Furthermore, they analyzed the interactions of hepatotoxic compounds with bioentities such as carriers, transporters, and metabolizing enzymes.…”

“…Specifically, most predictive models comprise hundreds to thousands of small chemistry datasets that cannot cover enough chemical space [76,118,188]. Moreover, the data is usually dispersed to many literatures [117,118,122,124,125,[128][129][130][134][135][136][137][138][139][140][141][142][143][144]150,155,[159][160][161][162][163][164][165][166][167][168][169]173,176,179,183], is unbalanced, and has cutoff ambiguity challenges [118,167]. Furthermore, the bioactivity assay data is strongly biased to its platform, and has an intrinsic experimental error which disrupts accurate prediction [189].…”

In silico Identification and Mechanism Exploration of Hepatotoxic Ingredients in Traditional Chinese Medicine

“…Liver injury induced by drug, novel foods or phytotherapy, also known as hepatotoxicity, is still a major clinical and pharmaceutical concern (Amadi and Orisakwe, 2018; Hammann et al, 2018; Kyawzaw et al, 2018; Real et al, 2018). According to the data from United States National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), hepatotoxicity accounts for 50% of all liver failure cases in the United States (Tujios and Fontana, 2011).…”