Machine learning (ML) systems—predictive computer programs that get better at spotting patterns in data through an iterative process that is similar to human learning—could solve biopharma’s most challenging process development and control problems.

However, realizing this potential will take effort, according to Ratnesh Jain, PhD, from the Institute of Chemical Technology in Mumbai, who says cutting production times, costs, and even drug prices with ML will require investment and innovation.

“In process development, ML methods can offer new insights into critical quality attributes (CQA) and critical process parameters (CPP), which can facilitate better process control. Supervised learning algorithms may expedite high-throughput screening experiments, such as clone selection, media screening, and feed development strategies, and resin and column dimension selection, formulation development, etc,” he says.

“ML could also help to predict deviations in product quality, assist in effective decision making, which in turn may enable real-time-release (RTR) of biotechnological products. Thus, ML-based methods for biopharmaceutical process development and manufacturing may reduce the time and cost involved therein and provide access to affordable medicines.”

Jain cites things like handling missing information in upstream datasets or the optimization of purification sequence and chromatographic column sizing strategies during therapeutic monoclonal antibody production as examples where ML can make a difference on the manufacturing line.

Analysis and prediction of post-translational modifications

Jain also points out the ML’s potential for use in the analysis and prediction of post translational modifications (PTMs), including those changes to therapeutic proteins that are associated with clinical efficacy and safety. Turning this potential into monitoring and control systems that are ready to use on the factory floor is the tricky part as, according to a recent study by Jain and colleagues, industrial ML is in its infancy.

“Deriving the vast benefits from ML in biopharmaceutical development needs commitment and involvement from multiple stakeholders across biotechnology and other technology-based companies, regulators, policy makers, academic institutes, hospitals, and healthcare providers,” he tells GEN.”

And there is some good news. In recent years influential regulators have started planning for machine learning based production Jain says, citing the FDA as an example.

“Regulatory authorities are setting new pathways for evaluation and adoption of AI and ML in healthcare,” continues Jain. “Signing of the ‘21st century cure act’ has been regarded as a substantial bipartisan legislative achievement targeted at expediting the discovery, development, and delivery of innovative cures and treatments.”

But even with regulatory support, drug companies will still need infrastructure to make use of ML in commercial production, according to Jain, who says an internet of things (IOT) strategy should be a standard part of the facility design process.

“Application of IOT in biopharmaceutical manufacturing plants to enable real-time data acquisition, inter-operability, and intelligent communication will help to effectively implement ML at commercial scale,” Jain explains. “The basic requirement for making use of ML for prediction and control is availability of suitable quality and quantity of data. Thus, the manufacturer must look for implementation of advanced instruments with ability to capture real-time data, computing ability, and ability to inter-communicate among machines.”

Ultimately though, Jain predicts efforts by industry, the tech sector, regulators, and computer scientists will work and help biopharma make the most of machine learning.

“With the effective incorporation of inter-disciplinary fields such as quantum computing, IOT, quantum cryptography, digitization, etc., we may foresee ML to be commonly used in biopharma in the upcoming five to ten years,” he says.