Research interests and contributions

Polymer structure predictor (PSP): A python toolkit for generating a hierarchy of polymer models

A 3D atomic-level polymer model is the starting point for any physics-based simulation study, and an autonomous model generator is in high demand. We developed a python toolkit named Polymer Structure Predictor (PSP) for building a hierarchy of polymer models, ranging from oligomers to the infinite polymer chain, crystal and amorphous structures. The only input of PSP is the simplified molecular-input line-entry system (SMILES) strings of the polymer repeating unit. The output structures and accompanying force field (GAFF2/OPLS-AA) parameter files can be used for downstream in-silico simulations with a variety of computational packages, including VASP, ORCA, LAMMPS, and GAMESS. To ensure the reliability of polymer models, they were compared with known structures, and various properties of polymers, such as bandgap, ring-opening polymerization enthalpy, and glass transition temperature, were computed using ab initio/MD simulations and verified against reported experimental data. PSP has already been integrated with the polymer version of computational autonomy for materials discovery (CAMD)[Chem. Sci., 2020, 11, 8517], an artificial intelligence-based platform for discovering polymers with targeted performance. The PSP package includes a Colab notebook using which users can go through several examples, building their own models, visualizing them, and downloading them for later use. This open-source package is distributed under the MIT license and can be obtained from GitHub at https://github.com/Ramprasad-Group/PSP.

An informatics approach for designing conducting polymers

Conductive polymers are highly desired for many applications, such as thermoelectrics, transparent electrodes, and hole/electron transport layers. It is still challenging to select a polymeric material with a conductivity of 1000 S/cm, which is a major drawback for further development of electronic devices. To realize metal-like conductivity, I built machine-learning (ML) models using experimental data and three hierarchical levels of descriptors, i.e., (1) atomic-level fragments, (2) block-level fragments, and (3) chain-level features. Nearly 845,000 new candidates are screened, and promising candidates are identified for their potential applications in organic devices. New guidelines are proposed for designing conducting polymers by performing Shapley additive explanations (SHAP) analysis and computing the importance (z-score) of molecular fragments. Models are also incorporated at www.polymergenome.org to provide easy access for predicting the electrical conductivity of desired polymer/dopant blends.

Designing organic materials for solar cell applications using machine learning

Designing promising molecules is in high demand for further increasing the efficiency of organic photovoltaics (OPVs), which can be accelerated through virtual screening of a large number of possible candidate molecules. The Scharber model, widely used in current virtual screening works for OPVs, is not accurate enough (r~0.2) for the prediction of power conversion efficiency (PCE) as it is based on only energies of the frontier molecular orbitals of organic molecules. An efficient and computationally cheap model is desired to explore the chemical space of conjugated molecules. However, building a machine learning (ML) model is highly challenging as the complete set of descriptors required to capture the complexity of the energy conversion process is still unknown. In this project, we first established a sufficiently large dataset for experimental PCE of reported molecules with a wide variety of chemical structures. Starting from introducing a set of relevant descriptors, we trained several ML models using gradient boosting regression tree (GBRT) and artificial neural network (ANN) algorithms by 10-fold cross-validation technique in combination with the stratified sampling method, and impressive performance (r=0.8) is achieved for a GBRT model.

High-throughput virtual screening of ~10,000 candidate molecules, constructed from commonly used 32 unique building blocks, are performed using these ML models. Important building blocks are identified, and new design rules are introduced to construct efficient molecules. Combining the results of both GBRT and ANN models, 126 candidates with theoretically predicted efficiency > 8% are proposed for further analysis, synthesis, and device fabrication.

Although predicting the PCE using our ML models is quite promising, it is highly desired to improve a specific device property to fulfill a particular commercial requirement. For example, solar-to-fuel energy conversion requires high open-circuit voltage (Voc) whereas a high short circuit current (Jsc) is necessary for solar window applications. A series of machine learning models are built for three important device characteristics (Voc, Jsc, and fill factor) using RF and GBRT algorithms and relevant molecular properties as descriptors, resulting in an impressive predictive performance (r=0.7). These models may play a vital role in designing promising organic materials for a specific photovoltaic application with high Voc/Jsc. Important descriptors for each device parameter are identified, and their relations to the energy conversion process are discussed in detail. Notably, the energetic differences of lowest occupied molecular orbital (LUMO) and LUMO+1 of a donor molecule (∆L), the energetic difference of LUMO of donor and LUMO of acceptor, and change in dipole moment in going from the ground state to the first excited state for donor molecules (∆μge), not frequently considered for designing OPV materials, are found to be important for tuning device parameters.

In-silico investigation of optical and electronic properties of helical oligomers

In addition to being of immense biological interest, helices are progressively seeking attention for applications in materials science. Their unique helical structures fulfill many essential needs. If one of the two helices (i.e., right and left-handed helices) can be selectively synthesized, all helical oligomers will behave as optically active material due to their inherent chiral structures. A helical π-conjugated polymer having high conductivity is also suitable for applications like a molecular solenoid. Our study on the stabilities and optoelectronic properties of different conformers of π-spaced heterocyclic oligomers shows that helical conformers of vinylene-linked systems are the most stable conformers, and their properties are quite different than their respective linear conformers. A detailed study on pyridine-furan, pyridine-pyrrole, and pyridine-thiophene oligomers shows that their helical conformers are feasible. The absorption spectra of these oligomers are composed of multiple electronic transitions having significant oscillator strengths.

In-silico investigation of optical and electronic properties of heterocyclic conjugated polymers

Heterocyclic conjugated polymers are extensively studied for their applications in organic devices, such as organic light-emitting diodes, organic photovoltaics, and organic field-effect transistors. Lack of appropriate optical gap, improper alignment of frontier orbitals with respect to the work function of electrodes, small mobilities of charge carriers, inappropriate hole-electron binding energy, etc., of organic materials often limit the performance of organic devices compared to the inorganic devices. Although many studies have been devoted to finding means and methods to improve the performance of these organic devices, designing an appropriate material with the required electronic and optical properties is still a challenging task. Computational studies, at this juncture, are of immense help in understanding the physical properties of existing conjugated polymers and designing new materials. The objective of this project was to explore the structure and properties of various heterocyclic conjugated polymers using density functional theory (DFT) and time-dependent DFT (TDDFT) methods. Our study reveals that a few heterocyclic azomethines are suitable for applications such as light absorption, hole injection, and hole transport in organic devices.