Bioplastic design using multitask deep neural networks

Abstract

Non-degradable plastic waste jeopardizes our environment, yet our modern lifestyle and current technologies are impossible to sustain without plastics. Bio-synthesized and biodegradable alternatives such as polyhydroxyalkanoates (PHAs) have the potential to replace large portions of the world’s plastic supply with cradle-to-cradle materials, but their chemical complexity and diversity limit traditional resource-intensive experimentation. Here, we develop multitask deep neural network property predictors using available experimental data for a diverse set of nearly 23,000 homo- and copolymer chemistries. Using the predictors, we identify 14 PHA-based bioplastics from a search space of almost 1.4 million candidates which could serve as potential replacements for seven petroleum-based commodity plastics that account for 75% of the world’s yearly plastic production. We also discuss possible synthesis routes for the identified promising materials.

Informatics-Driven Selection of Polymers for Fuel-Cell Applications

“…About 98.5% (2,105 entries) of the proton conductivity and 42.6% (801 entries) of the water uptake data set involve the ionic −SO 3 – sulfonate group. From the data learning standpoint, data sets of correlated properties can be fused and learned simultaneously in a multitask (MT) ML model so that possible hidden correlations among them can be accessed. , Likewise, a data set of 2624 data points for the permeabilities of six gases, including H 2 , O 2 , He, CO 2 , N 2 , and CH 4 , were taken from previous works, , augmented, and learned in another MT model. Four other data sets of polymer band gap E g , thermal decomposition temperature T d , glass transition temperature T g , and Young’s modulus E were also utilized from past works. ,, We note that, among these data sets, only those for σ and λ contain the temperature T , the relative humidity RH, and information on if the water is in the liquid form or not, when the measurements were made.…”

“…The screening space contains 30 624 known polymers, including 16 858 copolymers and 13,766 homopolymers, all of them have been synthesized and reported in the literature. Some more information on this data set can be found in refs , , and . In addition to the smiles strings and respective concentrations, references pointing to the report of each polymer are also available.…”

“…Four other data sets of polymer band gap E g , thermal decomposition temperature T d , glass transition temperature T g , and Young's modulus E were also utilized from past works. 40,41,85 We note that, among these data sets, only those for σ and λ contain the temperature T, the relative humidity RH, and information on if the water is in the liquid form or not, when the measurements were made. Such information is unavailable in the other data sets.…”

Emerging Trends in Machine Learning: A Polymer Perspective

“…As previously stated, one of the ultimate goals of ML for polymers is materials/process design. This often takes the form of optimization, most commonly property optimization. − ,,, It is important to note that while most efforts have focused on optimizing the chemistry or formulation, one can also optimize the materials processing steps (e.g., annealing, film casting, mixing conditions). One can also simultaneously optimize multiple quantities. , Materials optimization falls into the category of inverse design where the goal is to find an input (e.g., synthesis or processing parameters) that yields a desired output (e.g., material property).…”

“…Over the past five years or so, there has been tremendous progress in the application of these methods to polymer problems as detailed in numerous perspectives and reviews. − Polymers focused researchers are using ML to accelerate the discovery of new materials and new knowledge, as well as working to overcome barriers such as data scarcity. For example, ML has enabled the generation of potential new polymer chemistries, new materials for gas separation membranes, prediction of properties for sequence defined polymers, bioplastic design, guidance for improving 3D printing, improved contrast agents for magnetic resonance imaging (MRI) measurements, and methods for improved predictions of very small data sets …”