Graph Machines and Their Applications to Computer-Aided Drug Design: A New Approach to Learning from Structured Data. In: Calude CS, Dinneen MJ, Păun G, Rozenberg G, Stepney S, editors. Unconventional Computation. Berlin: Springer Berlin Heidelberg; 2006. pp. 1-19. [DOI: 10.1007/11839132_1]

Fast Prediction of the Equivalent Alkane Carbon Number Using Graph Machines and Neural Networks

The hydrophobicity of oils is a key parameter to design surfactant/oil/water (SOW) macro-, micro-, or nano-dispersed systems with the desired features. This essential physicochemical characteristic is quantitatively expressed by the equivalent alkane carbon number (EACN) whose experimental determination is tedious since it requires knowledge of the phase behavior of the SOW systems at different temperatures and for different surfactant concentrations. In this work, two mathematical models are proposed for the rapid prediction of the EACN of oils. They have been designed using artificial intelligence (machine-learning) methods, namely, neural networks (NN) and graph machines (GM). While the GM model is implemented from the SMILES codes of a 111-molecule training set of known EACN values, the NN model is fed with some σ-moment descriptors computed with the COSMOtherm software for the 111-molecule set. In a preliminary step, the leave-one-out algorithm is used to select, given the available data, the appropriate complexity of the two models. A comparison of the EACNs of liquids of a fresh set of 10 complex cosmetic and perfumery molecules shows that the two approaches provide comparable results in terms of accuracy and reliability. Finally, the NN and GM models are applied to nine series of homologous compounds, for which the GM model results are in better agreement with the experimental EACN trends than the NN model predictions. The results obtained by the GMs and by the NN based on σ-moments can be duplicated with the demonstration tool available for download as detailed in the Supporting Information.

The relationship between cognition and vocational training outcome in patients with acquired brain injury: Contribution of Machine Learning

The aim of this study was to identify specific cognitive patterns related to long-term vocational training outcome. Records of twenty-eight patients who had benefited from a professional rehabilitation program were retrospectively processed. Screening through machine learning algorithms of patients' neuropsychological scores identified cognitive patterns related to both vocational training outcomes: succeeded or failed. These patterns were based on cognitive performance intervals even if the cognitive ability was not impaired. The cognitive pattern related to a successful vocational training included performance intervals on measures underlying verbal memory consolidation, visual memory incidental recall, problem solving and planning abilities. The cognitive pattern explaining failure of vocational training included performance intervals on tasks involving planning and problem solving abilities. From a comprehensive neuropsychological battery, memory and executive measures appeared to be the best attributes related to the vocational training outcome in patients with brain injury. Even with a cognitive functioning above the pathological cutoff, patients suffering from a brain injury could fail a vocational training. The attributes related to the vocational training outcome would be more a specific level of cognitive functioning rather than an interpretation of neuropsychological scores only based on a normal versus pathological distinction.

Machine learning methods for property prediction in chemoinformatics: Quo Vadis?

This paper is focused on modern approaches to machine learning, most of which are as yet used infrequently or not at all in chemoinformatics. Machine learning methods are characterized in terms of the "modes of statistical inference" and "modeling levels" nomenclature and by considering different facets of the modeling with respect to input/ouput matching, data types, models duality, and models inference. Particular attention is paid to new approaches and concepts that may provide efficient solutions of common problems in chemoinformatics: improvement of predictive performance of structure-property (activity) models, generation of structures possessing desirable properties, model applicability domain, modeling of properties with functional endpoints (e.g., phase diagrams and dose-response curves), and accounting for multiple molecular species (e.g., conformers or tautomers).

Molecular evolution of a peptide GPCR ligand driven by artificial neural networks

Peptide ligands of G protein-coupled receptors constitute valuable natural lead structures for the development of highly selective drugs and high-affinity tools to probe ligand-receptor interaction. Currently, pharmacological and metabolic modification of natural peptides involves either an iterative trial-and-error process based on structure-activity relationships or screening of peptide libraries that contain many structural variants of the native molecule. Here, we present a novel neural network architecture for the improvement of metabolic stability without loss of bioactivity. In this approach the peptide sequence determines the topology of the neural network and each cell corresponds one-to-one to a single amino acid of the peptide chain. Using a training set, the learning algorithm calculated weights for each cell. The resulting network calculated the fitness function in a genetic algorithm to explore the virtual space of all possible peptides. The network training was based on gradient descent techniques which rely on the efficient calculation of the gradient by back-propagation. After three consecutive cycles of sequence design by the neural network, peptide synthesis and bioassay this new approach yielded a ligand with 70fold higher metabolic stability compared to the wild type peptide without loss of the subnanomolar activity in the biological assay. Combining specialized neural networks with an exploration of the combinatorial amino acid sequence space by genetic algorithms represents a novel rational strategy for peptide design and optimization.

Predicting activities without computing descriptors: graph machines for QSAR

We describe graph machines, an alternative approach to traditional machine-learning-based QSAR, which circumvents the problem of designing, computing and selecting molecular descriptors. In that approach, which is similar in spirit to recursive networks, molecules are considered as structured data, represented as graphs. For each example of the data set, a mathematical function (graph machine) is built, whose structure reflects the structure of the molecule under consideration; it is the combination of identical parameterised functions, called "node functions" (e.g. a feedforward neural network). The parameters of the node functions, shared both within and across the graph machines, are adjusted during training with the "shared weights" technique. Model selection is then performed by traditional cross-validation. Therefore, the designer's main task consists in finding the optimal complexity for the node function. The efficiency of this new approach has been demonstrated in many QSAR or QSPR tasks, as well as in modelling the activities of complex chemicals (e.g. the toxicity of a family of phenols or the anti-HIV activities of HEPT derivatives). It generally outperforms traditional techniques without requiring the selection and computation of descriptors.