Inverse Molecule Design with Invertible Neural Networks as Generative Models

Using neural networks for supervised learning means learning a function that maps input x to output y. However, in many applications, the inverse learning is also wanted, i.e., inferring y from x, which requires invertibility of the learning. Since the dimension of input is usually much higher than that of the output, there is information loss in the forward learning from input to output. Thus, creating invertible neural networks is a difficult task. However, recent development of invertible learning techniques such as normalizing flows has made invertible neural networks a reality. In this work, we applied flow-based invertible neural networks as generative models to inverse molecule design. In this context, the forward learning is to predict chemical properties given a molecule, and the inverse learning is to infer the molecules given the chemical properties. Trained on 100 and 1000 molecules, respectively, from a benchmark dataset QM9, our model identified novel molecules that had chemical property values well exceeding the limits of the training molecules as well as the limits of the whole QM9 of 133,885 molecules, moreover our generative model could easily sample many molecules (x values) from any one chemical property value (y value). Compared with the previous method in the literature that could only optimize one molecule for one chemical property value at a time, our model could be trained once and then be sampled any multiple times and for any chemical property values without the need of retraining. This advantage comes from treating inverse molecule design as an inverse regression problem. In summary, our main contributions were two: 1) our model could generalize well from the training data and was very data efficient, 2) our model could learn bidirectional correspondence between molecules and their chemical properties, thereby offering the ability to sample any number of molecules from any y values. In conclusion, our findings revealed the efficiency and effectiveness of using invertible neural networks as generative models in inverse molecule design.

SYNTHESIS OF MBS RESIN BASED ON MOLECULE DESIGN

The relationships between the properties of PVC-MBS polyblend and of the MBS multiphase structure and between the MBS structure and polymerization parameters are studied theoretically and experimentally. MBS resin synthesized from the molecule design principle is suitable to prepare PVC-MBS polyblend with good transparency and high impact strength.

Reinforcement Learning of Molecule Optimization with Bayesian Neural Networks

Creating new molecules with desired properties is a fundamental and challenging problem in chemistry. Reinforcement learning (RL) has shown its utility in this area where the target chemical property values can serve as a reward signal. At each step of making a new molecule, the RL agent learns selecting an action from a list of many chemically valid actions for a given molecule, implying a great uncertainty associated with its learning. In a traditional implementation of deep RL algorithms, deterministic neural networks are typically employed, thus allowing the agent to choose one action from one sampled action at each step. In this paper, we proposed a new strategy of applying Bayesian neural networks to RL to reduce uncertainty so that the agent can choose one action from a pool of sampled actions at each step, and investigated its benefits in molecule design. Our experiments suggested the Bayesian approach could create molecules of desirable chemical quality while maintained their diversity, a very difficult goal to achieve in machine learning of molecules. We further exploited their diversity by using them to train a generative model to yield more novel drug-like molecules, which were absent in the training molecules as we know novelty is essential for drug candidate molecules. In conclusion, Bayesian approach could offer a balance between exploitation and exploration in RL, and a balance between optimization and diversity in molecule design.

Group Contribution Method Supervised Neural Network for Precise Design of Organic Nonlinear Optical Materials

To rationalize the design of D-π-A type organic small-molecule nonlinear optical materials,a theory guided machine learning framework is constructed.Such an approach is based on the recognition that the optical property of the molecule is predictable upon accumulating the contribution of each component,which is in line with the concept of group contribution method in thermodynamics.To realize this,a Lewis-mode group contribution method(LGC)has been developed in this work,which is combined with the multistage Bayesian neural network and the evolutionary algorithm to constitute an interactive framework(LGC-msBNN-EA).Thus,different optical properties of molecules are afforded accurately and efficientlyby using only a small data set for training.Moreover,by employing the EA model designed specifically for LGC,structural search is well achievable.The origins of the satisfying performance of the framework are discussed in detail.Considering that such a framework combines chemical principles and data-driven tools,most likely,it will be proven to be rational and efficient to complete mission regarding structure design in related fields.

Exploring Local Chemical Space in De Novo Molecular Generation Using Multi-Agent Deep Reinforcement Learning

Single-agent reinforcement learning (RL) is commonly used to learn how to play computer games, in which the agent makes one move before making the next in a sequential decision process. Recently single agent was also employed in the design of molecules and drugs. While a single agent is a good fit for computer games, it has limitations when used in molecule design. Its sequential learning makes it impossible to modify or improve the previous steps while working on the current step. In this paper, we proposed to apply the multi-agent RL approach to the research of molecules, which can optimize all sites of a molecule simultaneously. To elucidate the validity of our approach, we chose one chemical compound Favipiravir to explore its local chemical space. Favipiravir is a broad-spectrum inhibitor of viral RNA polymerase, and is one of the compounds that are currently being used in SARS-CoV-2 (COVID-19) clinical trials. Our experiments revealed the collaborative learning of a team of deep RL agents as well as the learning of its individual learning agent in the exploration of Favipiravir. In particular, our multi-agents not only discovered the molecules near Favipiravir in chemical space, but also the learnability of each site in the string representation of Favipiravir, critical information for us to understand the underline mechanism that supports machine learning of molecules.

Effect of Molecular Structure on the Performance of Polyacrylic Acid Superplasticizer

The effects of structure parameters, such as molecular structure, segment kinds, molecular weight, and organic functional groups, on the performance of polyacrylic acid superplasticizer were discussed. According to the differences of chain sections, functional groups, eic, polyacrylic acid superplasticizer could be divided into A, B, C three parts. Among them, A chain section included sulfonic acid groups, B chain section carboxyl groups, C chain section polyester. Polyacrylic acid superplasticizers with different matching of A, B, C chain sections, different length of C chain section and different molecular weights were synthesized by acrylic acid, polyethylene glycol, sodium methyl allylsulfonate; the relation between the molecular structure and perfolxnance was also studied. The expetimental results indicate that the water-reduction ratio increases obviously with the increment of the proportion of sodium methyl allylsulfonate chain section in the molecular; the slump retention increases greatly with the increment of the proportion of acrylic acid chain section; the dispersion of cement particles increases with the increment of the chain length of polyethylene glycol; when the molecular weight is in the range of 5000, the dispersion and slump retentibity increase with the increment of the average molecular weight of polymers.

Design of solvent mixtures for removal of phenol from wastewater using a non-linear programming model with a multi-start method

The optimization-based design of solvent mixtures used for phenolic wastewater treatment was investigated in this work.A nonlinear programming(NLP)model was formulated based on the concepts of computer-aid molecule design(Computer-Aided Molecular Design,CAMD)to select solvent mixtures with the best extraction performance considering the constraints of extraction process and the environmental impact.Due to the complexity of the NLP model,multi-start method was adopted to solve this problem in order to get near global optimal solution.The results of the calculations suggested that the optimal mixture consisted of 70.1%n-octanol and 29.9%2-octanone(molar fraction).The 119 sets of experimental results showed that the extraction ability of the optimal solvent mixture identified by CAMD technique was among the top 6 sets compared to the experiment results.The results suggested that the developed NLP model could be able to screen the optimal solvent mixture in phenolic wastewater treatment.

D-A structural protean small molecule donor materials for solution-processed organic solar cells

Under the synergistic effect of molecular design and devices engineering, small molecular organic solar cells have presented an unstoppable tendency for rapid development with putting forward donor- acceptor （D-A） structures. Up to now, the highest power conversion efficiency of small molecules has exceeded 11%, comparable to that of polymers. In this review, we summarize the high performance small molecule donors in various classes of typical donor-acceptor （D-A） structures and discuss their relationships briefly.

Influence of the replacement of alkoxyl with alkylthienyl on photovoltaic properties of two small molecule donors for organic solar cells

Two benzo[1,2-b:4,5-b￠]dithiophene(BDT)-based small molecule(SM) donor materials with identical conjugated backbones but different substitution groups, named as DRTB-O and DRTB-T, were well explored to demonstrate the influence of the replacement of alkoxy with alkylthienyl on their photovoltaic properties in fullerene-based and fullerene-free organic solar cells(OSCs). The study shows that the two SM donors possess similar absorption spectra and energy levels but different crystalline structures in solid films. The carrier transport property and phase separation morphologies of the blend films have also been fully investigated.By employing PC71 BM as the acceptor, the power conversion efficiency(PCE) of DRTB-O:PC71BM and DRTB-T:PC71BM based devices were 4.91% and 7.08%, respectively. However, by blending with IDIC, the two SM donors exhibited distinctly different photovoltaic properties in fullerene-free OSCs, and the PCE of DRTB-O:IDIC and DRTB-T:IDIC based devices were 0.15% and9.06%, respectively. These results indicate that the replacement of alkoxyl with alkylthienyl in designing SM donor materials plays an important role in the application of fullerene-free OSCs.

Pinning-down molecules in their self-assemblies with multiple weak hydrogen bonds of C-H···F and C-H···N

Two-dimensional self-assemblies of four partially fluorinated molecules, 1,4-bis（2,6-difluoropyridin-4-yl）benzene, 4,4＇-bis（2,6-difluoropyridin-4-yl）-1,1＇-biphenyl, 4,4＇-bis（2,6-difluoropyridin-4-yl）-1,1＇：4＇,1＇-terphenyl and 4,4＇-bis（2,6-difluoropyridin-3-yl）-1,1＇-biphenyl, involving weak intermolecular C-H···F and C-H···N hydrogen bonds were systematically investigated on Au（111） with low-temperature scanning tunneling microscopy. The inter-molecular connecting modes and binding sites were closely related to the backbones of the building blocks, i.e., the molecule length determines its binding sites with neighboring molecules in the assemblies while the attaching positions of the N and F atoms dictate its approaching and docking angles. The experimental results demonstrate that multiple weak hydrogen bonds such as C-H···F and C-H···N can be efficiently applied to tune the molecular orientations and the self-assembly structures accordingly.

X-site doping in ABX_(3) triggers phase transition and higher T_(c) of the dielectric switch in perovskite

Material stability is always the key factor for applied materials especially the working environment that requires higher temperature sensitivity or temperature fluctuation range.In which,the stimulus-response perovskite materials are just sensitive to stability to ensure the accuracy and stability of the signals,in the applied devices of batteries and memory storage devices and so on.However,it is still a tremendous challenge to improve the stability of perovskite materials,and maintain reliability in the devices.Here,a novel ABX_(2)X'_(1)(X-site doping in an ABX_(3))compound[CEMP]-[CdBr_(2)(SCN)](1,CEMP=1-(2-chloro-ethyl)-1-methyl-piperidine)with remarkable high-temperature reversible dielectric switching behavior was proposed.The strategy of[SCN]^(−)doping in perovskite for improving the stability was successfully achieved.Meanwhile,the steric hindrance is increased while the energy barrier is also increased by replacing hydrogen with flexible groups,which leads to a high-temperature reversible phase transition.The new finding provides a new direction to enrich new applications and design ideas of perovskite materials.Especially the X-site strategy of doping or substitution in the ABX_(3),it will promote ingenious and perfect experimental results in material synthesis and performance improvement by chemistry disciplines.

Molecular Modeling Studies of 4-Hydroxyamino α-Pyranone Carboxamide Analogues as Hepatitis C Virus Inhibitor Using 3D-QSAR and Molecular Docking

In this paper, 42 4-hydroxyamino α-pyranone carboxamide analogues as Hepatitis C Virus(HCV) inhibitor 3 D-QSAR model was built based on Topomer CoMFA. The non-cross-validation(r2), cross-validation(q2), correlation coefficient of external validation(Q ext2), non-cross validated standard error(SD), standard error of prediction(SDCV) and F are 0.909, 0.615, 0.967, 0.13, 0.28 and 37.287, respectively. The obtained Topomer CoMFA model has good estimation stability and prediction capability. Topomer Search was employed as a tool for virtual screening in lead-like compounds in the ZINC database. Then, 6 R1 groups and 4 R2 groups with higher contribution values were employed to alternately substitute for the R1 and R2 of the template compound 21 with the highest bioactivity. As a result, 22 new molecules with higher activity than that of the template molecule were designed successfully. The Topomer Search technology could be effectively applied to screen and design new 4-hydroxyamino α-pyranone carboxamide analogues. The molecular docking method was also used to study the interactions of these drugs by docking the ligands into HCV active site, which revealed the likely bioactive conformations. This study showed extensive interactions between the 4-hydroxyamino α-pyranone carboxamide analogues and the active sites of HCV(residues TYR466, GLN384, TYR383 and ASP335). The design of potent new inhibitors of HCV can get useful insights from these results.