This more accurate prediction can be visually observed too, but the other metrics (R2, Rp, and Rs) do not capture this increase in accuracy due to substantially different ComboScore variance between the compared test sets. of models, and 5 types of chemical features. The application of a powerful, yet uncommonly used, RF-specific technique for reliability prediction is also investigated. The evaluation of these models shows that it is possible to forecast the synergy of unseen drug mixtures with high accuracy (Pearson correlations between 0.43 and 0.86 depending on the considered cell collection, with XGBoost providing slightly better predictions than RF). We have also EC0489 found that restricting to the most reliable synergy predictions results in at least SQSTM1 2-fold error decrease with respect to employing the best learning algorithm without any reliability estimation. Alkylating providers, tyrosine kinase inhibitors and topoisomerase inhibitors are the medicines whose synergy with additional partner medicines are better expected by the models. Despite its leading size, NCI-ALMANAC comprises an extremely small part of all conceivable mixtures. Given their accuracy and reliability estimation, the developed models should drastically reduce the number of required tests by predicting which of the regarded as combinations are likely to be synergistic. prediction methods. Quantitative Structure-Activity Relationship (QSAR) models establish a mathematical relationship between the chemical structure of a molecule, encoded as a set of structural and/or physico-chemical features (descriptors), and its biological activity on a target. Such methods have been successfully used in a wide variety of pharmacology and drug design projects (Cherkasov et al., 2014), including malignancy study (Chen et al., 2007; Mullen et al., 2011; Ali and Aittokallio, 2018). QSAR models are traditionally built using simple linear models (Sabet et al., 2010; Pick et al., 2011; Speck-Planche et al., 2011, 2012) to forecast the activity of individual molecules against a molecular target. In the last 15 years, non-linear machine learning methods, such as Neural Network (NN) (Gonzlez-Daz et al., 2007), Support Vector Machine (SVM) (Doucet et al., 2007) or Random Forest (RF) (Singh et al., 2015), have also been used to create QSAR models. More recently, QSAR modeling has also accomplished accurate prediction of compound activity on non-molecular targets such as malignancy cell lines (Kumar et al., 2014). To extend QSAR modeling beyond individual molecules, the set of features from each molecule in the combination must be built-in. Various ways exist to encode two or more molecules as a feature vector, e.g., SIRMS descriptors (Kuz’min et al., 2008) for properties of mixtures or the CGR approach for chemical reactions (de Luca et al., 2012). Demanding validation strategies for the producing models have been developed too (Muratov et al., 2012). The most common representation of a drug pair is, however, the concatenation of features from both molecules (Bulusu et al., 2016). On the other hand, modeling drug combinations requires the quantification of their synergy. Several metrics exist to quantify synergy (Foucquier and Guedj, 2015) (e.g., Bliss independence Bliss, 1939, Loewe additivity Chou and Talalay, 1984, Highest solitary agent approach Greco et al., 1995 or Chou-Talalay Method Chou, EC0489 2010). These are implemented in various commercial and publicly available software packages for the analysis of combination data, e.g., Combenefit (Di Veroli et al., 2016), CompuSyn (http://www.combosyn.com) EC0489 or CalcuSyn (http://www.biosoft.com/w/calcusyn.htm). One major roadblock in drug synergy modeling has been the lack of homogeneous data (i.e., datasets generated with the same assay, experimental conditions and synergy quantification). This has been, however, alleviated from the recent availability of large datasets from High-Throughput Screening (HTS) of drug combinations on malignancy cell lines. For instance, Merck offers released an HTS synergy dataset (O’Neil et al., 2016), covering mixtures of 38 medicines and their activity against 39 malignancy cell lines (more than 20,000 measured synergies). This dataset has been used to build predictive regression and classification models using multiple machine learning methods (Preuer et al., 2018). AstraZeneca carried out a screening study, spanning 910 drug mixtures over 85 malignancy cell lines (over 11,000 measured synergy scores), which was subsequently utilized for a Desire challenge (Li et al., 2018; Menden et al., 2019). Very recently, the largest publicly available malignancy drug combination dataset has been provided by the US National Malignancy Institute (NCI). This NCI-ALMANAC (Holbeck et al., 2017) tested over 5,000 mixtures of 104 investigational and authorized medicines, with synergies measured against 60 malignancy cell lines, leading to more.