Machine Learning Driven Optimization for High Precision Cellular Droplet Bioprinting

Abstract

Controlled volume microliter cell-laden droplet bioprinting is important for precise biologics deposition, reliably replicating 3D microtissue environments for building cell aggregates or organoids. To achieve this, we propose an innovative machine-learning approach to predict cell-laden droplet volumes according to input parameters. We developed a novel bioprinting platform capable of collecting high-throughput droplet images and generating an extensive dataset for training machine learning and deep learning algorithms. Our research compared the performance of three machine learning and two deep learning algorithms that predict droplet volume based on numerous bioprinting parameters. By adjusting bioink viscosity, nozzle size, printing time, printing pressure, and cell concentration as input parameters, we precisely could control droplet sizes, ranging from 0.1 to 50 microliter in volume. We utilized a hydrogel precursor composed of 5% gelatin methacrylate and a mixture of 0.5% and 1% alginate, respectively. Additionally, we optimized the cell bioprinting process using green fluorescent protein-tagged 3T3 fibroblast cells. These models demonstrated superior predictive accuracy and revealed the interrelationships among parameters while taking minimal time for training and testing. This method promises to advance the mass production of organoids and microtissues with precise volume control for various biomedical applications.