Molecular Design Problems Research Articles

Data‐driven integrated design of solvents and extractive distillation processes

AbstractAs property and process models with many variables need to be considered, integrated computer‐aided molecular and process design (CAMPD) problems are computationally expensive. An efficient CAMPD approach is proposed for the simultaneous design of solvents and extractive distillation (ED) processes based on a data‐driven modeling strategy. First, artificial neural network (ANN)‐based process models are trained to replace the physical models conventionally used in CAMPD. Subsequently, optimization is performed to maximize process performance, through which optimal solvent properties and corresponding optimal process parameters are obtained. Then, real solvents approximating the optimal property values are identified from a large solvent database. Rigorous simulations of the ED process are performed to evaluate the performance of the optimal solvents and corresponding process parameters. Further economic evaluation (6.11% lower annual cost compared to the benchmark process) and chemical hazard assessment confirm that acetylacetone is a promising solvent for the ED separation of 1‐butene from 1,3‐butadiene.

Enabling the direct solution of challenging computer-aided molecular and process design problems: Chemical absorption of carbon dioxide

The search for improved CO2 capture solvents can be accelerated by deploying computer-aided molecular and process design (CAMPD) techniques to explore large molecular and process domains systematically. However, the direct solution of the integrated molecular-process design problem is very challenging as nonlinear interactions between physical properties and process performance render a large proportion of the search space infeasible. We develop a methodology that enables the direct and reliable solution of CAMPD for absorption–desorption processes, using the state-of-the-art SAFT-γ Mie group contribution approach to predict phase and chemical equilibria. We develop new feasibility tests and show them to be highly efficient at reducing the search space, integrating them in an outer-approximation algorithm. The framework is applied to design an aqueous solvent and CO2 chemical absorption–desorption process, with 150 CAMPD instances across three case studies solved successfully. The optimal solvents are more promising than those obtained with sequential molecular design approaches.

Benchmarking a novel ensemble docking method.

Docking is an important tool in the kit of researchers and engineers working on molecular design and drug development problems. It is computationally efficient enough that it can be used on thousands or tens of thousands of structures with reasonable access to a cluster environment, and has the capacity to find binding poses and even estimate affinities. It has many well-known weaknesses, however. One significant weakness is that it does not properly account for conformational heterogeneity in receptors. This general weakness can produce incredibly poor results in cases where this heterogeneity changes the conformation of the binding site, since the receptor's ensemble can be mostly populated with states where the pocket of interest is closed (sterically occluded), as is the case with cryptic pockets.

Molecular Design Method Using a Reversible Tree Representation of Chemical Compounds and Deep Reinforcement Learning.

Automatic design of molecules with specific chemical and biochemical properties is an important process in material informatics and computational drug discovery. In this study, we designed a novel coarse-grained tree representation of molecules (Reversible Junction Tree; “RJT”) for the aforementioned purposes, which is reversely convertible to the original molecule without external information. By leveraging this representation, we further formulated the molecular design and optimization problem as a tree-structure construction using deep reinforcement learning (“RJT-RL”). In this method, all of the intermediate and final states of reinforcement learning are convertible to valid molecules, which could efficiently guide the optimization process in simple benchmark tasks. We further examined the multiobjective optimization and fine-tuning of the reinforcement learning models using RJT-RL, demonstrating the applicability of our method to more realistic tasks in drug discovery.

Improving de novo molecular design with curriculum learning

Reinforcement learning is a powerful paradigm that has gained popularity across multiple domains. However, applying reinforcement learning may come at the cost of multiple interactions between the agent and the environment. This cost can be especially pronounced when the single feedback from the environment is slow or computationally expensive, causing extensive periods of non-productivity. Curriculum learning provides a suitable alternative by arranging a sequence of tasks of increasing complexity, with the aim of reducing the overall cost of learning. Here we demonstrate the application of curriculum learning for drug discovery. We implement curriculum learning in the de novo design platform REINVENT, and apply it to illustrative molecular design problems of different complexities. The results show both accelerated learning and a positive impact on the quality of the output when compared with standard policy-based reinforcement learning. While reinforcement learning can be a powerful tool for complex design tasks such as molecular design, training can be slow when problems are either too hard or too easy, as little is learned in these cases. Jeff Guo and colleagues provide a curriculum learning extension to the REINVENT de novo molecular design framework that provides problems of increasing difficulty over epochs such that the training process is more efficient.

Generative Chemical Transformer: Neural Machine Learning of Molecular Geometric Structures from Chemical Language via Attention.

Discovering new materials better suited to specific purposes is an important issue in improving the quality of human life. Here, a neural network that creates molecules that meet some desired multiple target conditions based on a deep understanding of chemical language is proposed (generative chemical Transformer, GCT). The attention mechanism in GCT allows a deeper understanding of molecular structures beyond the limitations of chemical language itself which cause semantic discontinuity by paying attention to characters sparsely. The significance of language models for inverse molecular design problems is investigated by quantitatively evaluating the quality of the generated molecules. GCT generates highly realistic chemical strings that satisfy both chemical and linguistic grammar rules. Molecules parsed from the generated strings simultaneously satisfy the multiple target properties and vary for a single condition set. These advances will contribute to improving the quality of human life by accelerating the process of desired material discovery.

Generalizing the Discrete Gibbs Sampler-Based λ-Dynamics Approach for Multisite Sampling of Many Ligands.

In this work, the discrete λ variant of the Gibbs sampler-based λ-dynamics (d-GSλD) method is developed to enable multiple functional group perturbations to be investigated at one or more sites of substitution off a common ligand core. The theoretical framework and special considerations for constructing discrete λ states for multisite d-GSλD are presented. The precision and accuracy of the d-GSλD method is evaluated with three test cases of increasing complexity. Specifically, methyl → methyl symmetric perturbations in water, 1,4-benzene hydration free energies and protein-ligand binding affinities for an example HIV-1 reverse transcriptase inhibitor series are computed with d-GSλD. Complementary MSλD calculations were also performed to compare with d-GSλD's performance. Excellent agreement between d-GSλD and MSλD is observed, with mean unsigned errors of 0.12 and 0.22 kcal/mol for computed hydration and binding free energy test cases, respectively. Good agreement with experiment is also observed, with errors of 0.5-0.7 kcal/mol. These findings support the applicability of the d-GSλD free energy method for a variety of molecular design problems, including structure-based drug design. Finally, a discussion of d-GSλD versus MSλD approaches is presented to compare and contrast features of both methods.

A comparative study of multi-objective optimization methodologies for molecular and process design

The need to consider multiple objectives in molecular design, whether based on techno-economic, environmental or health and safety metrics is increasingly recognized. There is, however, limited understanding of the suitability of different multi-objective optimization (MOO) algorithm for the solution of such design problems. In this work, we present a systematic comparison of the performance of five mixed-integer non-linear programming (MINLP) MOO algorithms on the selection of computer-aided molecular design (CAMD) and computer-aided molecular and process design (CAMPD) problems. The five methods are designed to address the discrete and nonlinear nature of the problem, with the aim of generating an accurate approximation of the Pareto front. They include: a weighted sum approach without global search phases (SWS), a weighted sum approach with simulated annealing (WSSA), a weighted sum approach with multi level single linkage (WSML), the sandwich algorithm with MLSL (SDML) and the non dominated sorting genetic algorithm-II (NSGA-II). The algorithms are compared systematically in two steps. The effectiveness of the global search methods is evaluated with SWS, WSSA and WSML. WSML is found to be most effective and a comparative analysis of WSML, SDML and NSGA-II is then undertaken. As a test set of these optimization techniques, two CAMD and one CAMPD problems of varying dimensionality are formulated as case studies. The results show that the SDML provides the most efficient generation of a diverse set of Pareto points, leading to the construction of an approximate Pareto front close to exact Pareto front.

Integration of safety and health aspects in a simultaneous process and molecular design framework

The aspects of safety and occupational health have been included as part of the design criteria in many process and molecular design problems. However, the main design objectives of these works are the molecular properties and process variables that offer the optimized process objective. Safety and health criteria are not strongly emphasized in these problems, since they are often introduced as design constraints or are not evaluated in an extensive manner. In this research, a novel simultaneous process-molecular design framework by incorporating safety and occupational health aspects is presented to ensure the developed system does not pose much harm. The hazard level is evaluated with the utilization of safety and health indexes to examine chemical properties and process conditions that may bring about potential risks. The overall hazard level is quantified through the use of index scores. The proposed strategy is demonstrated on a gas-absorption case study to design a solvent with its corresponding process that has optimized property functionalities, process performance, and safety and health attributes. This framework allows user to identify the process or molecular factors that may contribute to high risk potential, so that the appropriate safety and health measures can be implemented in the designed process.

Computer-aided reaction solvent design considering inertness using group contribution-based reaction thermodynamic model

Solvents have been widely used in process manufacturing industries. When involved in liquid-phase organic synthesis reactions, solvents can reduce the activation energy of reactions between the reactants and the transition state through solvation effects. However, undesirable side reactions can also be performed between solvents and the reaction system (the reactants and products), which should be avoided for producing unnecessary byproducts in the reaction system. In this paper, an optimization-based methodology is proposed for inert reaction solvent design. In this method, first, a Group Contribution (GC)-based reaction thermodynamic model is developed to quantitatively identify the thermodynamic feasibility of side reactions between solvents and the reaction system. Then, the SMARTS (SMiles Arbitrary Target Specification)-based reaction generation algorithm is employed to generate possible side reactions between solvents and the reaction system, helping to integrate the developed GC-based reaction thermodynamic model with the Computer-Aided Molecular Design (CAMD) problem for designing inert reaction solvents through the formulation and solution of the Mixed-Integer Non-Linear Programming (MINLP) model. Due to the nonlinear equations in the MINLP model, a decomposition-based solution strategy is employed to solve the optimization problem. Finally, two case studies are presented to demonstrate the feasibility and effectiveness of the proposed optimization-based methodology for promising inert reaction solvent design.

Computer-aided molecular product-process design under property uncertainties – A Monte Carlo based optimization strategy

A methodology is presented to solve computer-aided molecular design (CAMD) and process design model problems under consideration of fluid property uncertainty. The uncertainties of the group contribution (GC) property prediction models are quantified for which asymptotic approximation of the covariance of parameter estimation errors is performed following a regression analysis. A Monte Carlo sampling technique generates GC factor samples within the respective uncertainties, which are evaluated separately as constraints to the CAMD optimization problem. The methodology is applied to identify working fluid candidates for an organic Rankine cycle used as waste heat recovery system in a marine diesel engine. CAMD under property uncertainties allows (1) identifying robust and more reliable molecules with respect to property uncertainties (conservative approach) and (2) enhancing the search space in order to find potentially globally optimal working fluids (optimistic approach). Suitable Hydrofluoroolefins (HFO) have been identified as potential working fluids for waste heat recovery.

Hierarchical Representation Based Constrained Multi-objective Evolutionary Optimisation of Molecular Structures

We propose an efficient algorithm to generate Pareto optimal set of reliable molecular structures represented by group contribution methods. To effectively handle structural constraints we introduce goal oriented genetic operators to the multi-objective Non-dominated Sorting Genetic Algorithm-II (NSGA-II). The constraints are defined based on the hierarchical categorisation of the molecular fragments. The efficiency of the approach is tested on several benchmark problems. The proposed approach is highly efficient to solve the molecular design problems, as proven by the presented benchmark and refrigerant design problems.

Building and deploying a cyberinfrastructure for the data-driven design of chemical systems and the exploration of chemical space

The use of modern data science has recently emerged as a promising new path to tackling the complex challenges involved in the creation of next-generation chemistry and materials. However, despite the appeal of this potentially transformative development, the chemistry community has yet to incorporate it as a central tool in every-day work. Our research program is designed to enable and advance this emerging research approach. It is centred around the creation of a software ecosystem that brings together physics-based modelling, high-throughput in silico screening and data analytics (i.e. the use of machine learning and informatics for the validation, mining and modelling of chemical data). This cyberinfrastructure is devised to offer a comprehensive set of data science techniques and tools as well as a general-purpose scope to make it as versatile and widely applicable as possible. It also emphasises user-friendliness to make it accessible to the community at large. It thus provides the means for the large-scale exploration of chemical space and for a better understanding of the hidden mechanisms that determine the properties of complex chemical systems. Such insights can dramatically accelerate, streamline and ultimately transform the way chemical research is conducted. Aside from serving as a production-level tool, our cyberinfrastructure is also designed to facilitate and assess methodological innovation. Both the software and method development work are driven by concrete molecular design problems, which also allow us to assess the efficacy of the overall cyberinfrastructure.

Integration of Fuzzy Analytic Hierarchy Process into multi-objective Computer Aided Molecular Design

In this paper, a novel Computer Aided Molecular Design (CAMD) framework is developed to solve multi-objective molecular design problems. CAMD can be formulated as a multi-objective optimisation problem when there are multiple target properties to be optimised simultaneously. A major obstacle faced by multi-objective CAMD problems is the difficulty in assigning weighting factors to the target properties, since the relative importance of these factors is not always defined. It is particularly difficult to compare target properties which belong to different categories, such as physicochemical, safety, health and environmental properties, on a common scale. This paper presents a systematic CAMD algorithm built on Fuzzy Analytic Hierarchy Process (FAHP) to deal with the ambiguity involved in evaluating the weights of target properties in multi-objective CAMD problem. Instead of using exact numerical values, FAHP approach expresses the pairwise comparison of target properties through triangular fuzzy numbers, which allow the degree of confidence of decision maker to be quantified. Hence, the proposed approach can address the uncertainties arising from ambiguity involved during value judgement elicitation in multi-objective CAMD problems. The solutions generated provide a better balance of performance for a set of identified target properties. The proposed methodology is illustrated through a case study on designing a better solvent for extracting residual oil from palm pressed fibre.

A systematic methodology for multi-objective molecular design via Analytic Hierarchy Process

Abstract This paper presents a novel methodology for solving multi-objective Computer Aided Molecular Design (CAMD) problems. One of the major challenges in multi-objective CAMD problems is the subjectivity involved in assigning the weighting factors to each property that is optimised. It is difficult to define the relative importance of each property in CAMD problems as target properties that belong to different categories cannot be compared on a common scale. It is crucial to solve this issue as distinct solutions will be generated due to different weighting factors of each property. In this work, a systematic framework that combines CAMD and Analytic Hierarchy Process (AHP) is developed to deal with the ambiguities involved in assessing the relative importance weightings of target properties in multi-objective molecular design problem. Through AHP, molecular design problems can be approached as a hierarchical decision structure and different categories of target properties are combined in the same analysis. Besides, AHP has the ability to generate numerical priorities from the subjective knowledge in evaluating the relative importance of properties using pairwise comparison. A case study on solvent design for oil extraction from palm pressed fibre is solved to illustrate the proposed methodology.

Designing molecular complexes using free-energy derivatives from liquid-state integral equation theory

Complex formation between molecules in solution is the key process by which molecular interactions are translated into functional systems. These processes are governed by the binding or free energy of association which depends on both direct molecular interactions and the solvation contribution. A design goal frequently addressed in pharmaceutical sciences is the optimization of chemical properties of the complex partners in the sense of minimizing their binding free energy with respect to a change in chemical structure. Here, we demonstrate that liquid-state theory in the form of the solute–solute equation of the reference interaction site model provides all necessary information for such a task with high efficiency. In particular, computing derivatives of the potential of mean force (PMF), which defines the free-energy surface of complex formation, with respect to potential parameters can be viewed as a means to define a direction in chemical space toward better binders. We illustrate the methodology in the benchmark case of alkali ion binding to the crown ether 18-crown-6 in aqueous solution. In order to examine the validity of the underlying solute–solute theory, we first compare PMFs computed by different approaches, including explicit free-energy molecular dynamics simulations as a reference. Predictions of an optimally binding ion radius based on free-energy derivatives are then shown to yield consistent results for different ion parameter sets and to compare well with earlier, orders-of-magnitude more costly explicit simulation results. This proof-of-principle study, therefore, demonstrates the potential of liquid-state theory for molecular design problems.

Working fluid selection for organic Rankine cycles – Impact of uncertainty of fluid properties

This study presents a generic methodology to select working fluids for ORC (Organic Rankine Cycles) taking into account property uncertainties of the working fluids. A Monte Carlo procedure is described as a tool to propagate the influence of the input uncertainty of the fluid parameters on the ORC model output, and provides the 95%-confidence interval of the net power output with respect to the fluid property uncertainties. The methodology has been applied to a molecular design problem for an ORC using a low-temperature heat source and consisted of the following four parts: 1) formulation of process models and constraints 2) selection of property models, i.e. Peng–Robinson equation of state 3) screening of 1965 possible working fluid candidates including identification of optimal process parameters based on Monte Carlo sampling 4) propagating uncertainty of fluid parameters to the ORC net power output. The net power outputs of all the feasible working fluids were ranked including their uncertainties. The method could propagate and quantify the input property uncertainty of the fluid property parameters to the ORC model, giving an additional dimension to the fluid selection process. In the given analysis 15 fluids had an improved performance compared to the base case working fluid.

Homogenous multi-agent optimization for process systems engineering problems with a case study of computer aided molecular design

In this paper, we propose a novel homogenous multi-agent optimization (HMAO) framework for optimal design of large scale process system engineering problems. The platform is validated using a benchmark problems and a computer-aided molecular design (CAMD) problem. The molecular design problem is a solvent selection problem and it is formulated as a mixed integer nonlinear programming (MINLP) in which solute distribution coefficient of a candidate solvent is maximized subject to structural feasibility, thermodynamic property and process constraints. The model simultaneously determines the optimal decisions that include the size and the functional groups of the candidate solvents. In developing the HMAO framework, multiple efficient ant colony optimization (EACO) algorithms are considered as distinct algorithmic agents. We illustrate this approach through a real world case study of the optimal design of solvent for extraction of acetic acid from waste process stream using liquid–liquid extraction. The UNIFAC model based on the infinite dilution activity coefficient is used to estimate the mixture properties. The results show that quality of the objective function and the computational efficiencies are improved by a factor ranged from 1.475 to 4.137. The new solvents proposed in this work are with much better targeted thermodynamic properties compared to the solvents proposed so far in previous studies.

Robust design of optimal solvents for chemical reactions—A combined experimental and computational strategy

Solvents can have significant effects on chemical reactions, however, their huge number makes the selection very difficult and costly. This work presents a systematic method for the design of reaction solvents. Kinetic models are built by correlating experimentally determined reaction rate constants in a small set of known solvents with corresponding solvent theoretical descriptors determined from quantum chemical calculations. Optimal solvents are then identified from the solution of an optimization-based molecular design problem. Besides the deterministic optimization, a robust solvent design framework is proposed to identify solvents that possess the best reaction performance under model uncertainties. The methodology is exemplified for a competitive Diels–Alder reaction with the objective of maximizing the production of the desired product relative to that of the byproduct. Compared to the best experimentally identified solvent, a 10.9% improvement in reaction performance can be achieved with our designed solvent. It is proven that the proposed method is an efficient tool for fast identification of high-performance solvents for chemical reactions. Moreover, it potentially promotes the development of new solvents.

Evolutionary algorithm for de novo molecular design with multi-dimensional constraints

An evolutionary approach for solving molecular design problems with descriptors of varying dimensionality has been developed. Spatial fragment based descriptors are employed to generate candidate solutions within a population, which is evolved through the application of genetic operators toward an improved fitness. The candidate molecules are represented as graphs, and as such, customized operators of crossover and mutation have been developed to be compatible with this representation. The search space is conveniently represented through limitations on the occurrence of each fragment, as defined by the chosen data set, and the spatial capabilities of this space are captured through an initial conformational analysis. This spatial information is compressed and utilized to generate conformational space estimations throughout the algorithm, which expedites the search for solution graphs. The effect of various user determined input parameters is considered and exemplified through a case study involving the identification of solvents falling within a desired boiling point range, as estimated by a multi-dimensional property model.