Our Take
Computational screening doesn't replace CHO cell experiments—it reduces how many candidates you need to send to the lab, which shrinks both cost and cycle time but doesn't change the fundamental biology.
Why it matters
Drug makers face pressure to accelerate antibody development without sacrificing manufacturability. Early-stage computational filtering could compress preclinical timelines, but only if the in silico predictions hold up in real manufacturing. The gap is real: companies lack public data on what manufacturability looks like across failed programs.
Do this week
Bioprocess leads: map your current CHO screening bottleneck (how many mutations you test, how long selection takes, what you kill early) before evaluating in silico tools so you can measure whether computational pre-filtering actually saves lab days.
BigHat Biosciences Presented In Silico Antibody Screening at PEGS Boston
BigHat Biosciences, an AI company focused on protein engineering, demonstrated at PEGS Boston how computational models built on cell-free expression systems can predict antibody yields and biophysical properties before sending candidates to Chinese Hamster Ovary (CHO) cell culture.
Hunter Elliott, PhD, vice president of machine learning at BigHat, explained the workflow: "There's only so many experiments you can do by putting an antibody into CHO cells. With in silico tools augmenting that exploration side, we can build models that make predictions for improved sequences, screening many more antibodies in silico than we need to send to the lab."
The company's approach allows researchers to explore a wider range of potential mutations computationally, then select the handful with the highest predicted yields and improved biophysical properties for lab validation. Elliott described this as "derisking your processes because you're combining your experiments with the in silico tools you're using."
A key use case: starting with a suboptimal antibody candidate and iterating through several rounds of computational optimization before moving to physical experiments, rather than killing the molecule early based on initial screening.
Manufacturability Data Gaps Limit Model Confidence
The panel at PEGS Boston surfaced a critical limitation. Elliott noted that in silico models lack publicly available data on manufacturability and developability for drugs that failed to reach the clinic. Without negative examples, models cannot learn what makes antibodies difficult to produce at scale.
Concern also exists that computational screening could accidentally filter out the best-performing candidates. Elliott's response: predictive models make it easier to optimize from a suboptimal starting point, keeping imperfect sequences in the development loop longer rather than discarding them prematurely.
The real value proposition sits in cycle time and cost reduction, not discovery. If you can screen 100 mutations in silico and pick the 5 most promising for CHO validation, you save weeks of experiment design and execution. The trade-off is model accuracy. Early adoption requires tight feedback loops between preclinical teams and manufacturing to validate that in silico predictions hold up in production.
Build Internal Benchmarks Before Buying
If you manufacture antibodies or manage preclinical development, start by instrumenting your current screening process. How many mutation candidates do you generate per project? How many enter CHO culture? What is your cycle time from candidate design to manufacturability decision?
These metrics are your baseline. In silico tools will only show ROI if they reduce the number of CHO experiments or compress the timeline. Without a baseline, you cannot distinguish genuine acceleration from vendor marketing. Additionally, interrogate any vendor on training data: if their models learned from only clinical-stage programs, they are blind to the failure modes you most need to avoid.
Pair any computational tool adoption with tighter communication between preclinical research and manufacturing. Elliott emphasized this during the panel: success with harder-to-manufacture drugs depends on early handoff of manufacturability constraints to researchers designing sequences. In silico screening only works if manufacturing can articulate what "manufacturability" means to the model.