Although I obtained my Bachelor degree in chemistry from the Lomonosov Moscow State University of Fine Chemical Technologies working on the organic synthesis of theranostic agents, “real science” began for me when I joined the SCAMT Laboratory at ITMO University in St Petersburg to begin studying for my Master Degree. My research project, under the guidance of Professor Vladimir Vinogradov , involved nanomedicine, nanotechnology and molecular biology, which were completely new domains for me. While navigating this steep learning curve, I received one of the “Best Poster Awards” while participating in the 20th International Sol-Gel Conference in St Petersburg and published my very first article in ChemNanoMat with the help of Dr Artur Prilepskii and Professor Vladimir Vinogradov. While writing a second article, I learned how a polished, “carefully-considered-tothe-last-comma” research article should appear, and how difficult it is to have papers accepted for publication in high-impact journals, such as Chemical Communications.
My PhD research in SCAMT has followed my now-established tradition of working outside of my comfort zone, this time on the application of artificial intelligence (AI) and machine learning (ML) in materials science. The main goal of my PhD study is to show the possibility of AI-assisted inverse design of nanomaterials and drug delivery systems with a desired set of properties, progressing from the defined properties to the synthesis route. This is the opposite of the conventional approach, where a material is obtained and then one tries to find out what useful properties it has. As a part of this aspiration, I am attempting to train a neural network to propose the synthesis route of the material (initially calcium carbonate, to demonstrate the proof-of-concept) based on a simple hand drawing of how the material might appear under the scanning electron microscope (SEM).
The model is trained using a large dataset of nano- and microparticles of various shapes and sizes synthesized in our lab, by correlating the contours obtained from SEM images and the corresponding synthesis routes. This approach is potentially able to predict the synthesis route of materials with tunable shapes, sizes, surface morphologies and polydispersity. At present, I have a curated dataset of 250 nano- and microparticles, including their SEM contours and synthesis procedures using varying temperatures, concentrations, stirring rates, synthesis time, additives (e.g. polymers and surfactants), solvent systems etc. In this manner, we cover the overall condition space and all of the possible parameters influencing the crystal growth. Software coded in Python is used to detect particle contours on SEM images and construct the neural network core, with the next step being model training and hypothesis testing.
I dream about an era of materials design where we will set the properties for a particular application and only then invest resources in synthesis, and I hope that my research will bring us closer to this dream.