Bioplastic design using multitask deep neural networks

Non-degradable plastic waste stays for decades on land and in water, jeopardizing our environment; yet modern lifestyle current technologies are impossible to sustain without plastics. Bio-synthesized biodegradable alternatives such as the polymer family of polyhydroxyalkanoates (PHAs) have potential replace large portions world's supply with cradle-to-cradle materials, but their chemical complexity diversity limit traditional resource-intensive experimentation. In this work, we develop multitask deep neural network property predictors using available experimental data a diverse set nearly 23000 homo- copolymer chemistries. Using predictors, identify 14 PHA-based bioplastics from search space almost 1.4 million candidates which could serve replacements seven petroleum-based commodity plastics that account 75% yearly production. We discuss possible synthesis routes these identified promising materials. The developed made part Polymer Genome project at https://PolymerGenome.org.