Every few months, a paper crosses our desks with scientific breakthroughs that couldn't have been run the way they were run ten years ago. Not because the science changed, but because the infrastructure did. Liquid handlers, high-throughput screening (HTS) platforms, and automated design-make-test-analyze (DMTA) or design-build-test-learn (DBTL) loops have quietly become as load bearing to modern discovery as the assay itself.
Here are five recent peer-reviewed examples where the automation wasn't just a footnote in the methods section; it was the reason the results exist at all.
A Phage Cocktail that Survived Contact with 356 Clinical Isolates
Researchers at Locus Biosciences needed to find a fixed combination of bacteriophages capable of treating urinary tract infections (UTIs) across a genuinely diverse population of E. coli, not just the handful of lab strains most phage studies rely on. The math alone made this hard: testing every possible combination from a 500-phage collection against a 6-phage cocktail size would require more than 10 trillion combinations. The team ultimately identified LBP-EC01, a cocktail active against 96.4% of 356 clinically derived isolates, now in Phase 2 clinical trials.
Two liquid handling work cells, each combining 8-channel and 96-channel pipetting with shaking incubators and integrated plate readers, coordinated to execute more than 3.8 million phage-bacteria reactions. Computer vision replaced human colony counting almost entirely; the automated enumeration pipeline was accurate in 99.5% of samples, compared to 94-97% for trained human analysts. Without that combination of liquid handling throughput and automated imaging, a screen of this scale simply isn't a weekend project; it isn't a project at all.
An Antimicrobial Polymer Library that Outperformed a Clinical Antibiotic
A research team at Zhejiang University set out to design AMP-mimicking antimicrobial polymers, synthetic stand-ins for antimicrobial peptides that are cheaper to produce and more stable than the natural molecules. The chemical space was too large to explore by hand: 13,728 possible monomer and ratio combinations. Their winning candidate matched the in vivo efficacy of ceftazidime, a frontline clinical antibiotic, while remaining active against MRSA and Pseudomonas aeruginosa.
An automated liquid handling platform running programmed PET-RAFT polymerization steps, paired with a Design-Build-Test-Learn machine learning loop. The numbers tell the throughput story on their own: 400 polymer combinations that took roughly 120 hours by hand were synthesized and characterized in 10 hours automated, a 12-fold speedup, with the platform ultimately narrowing 13,728 candidates to 7 leads in 12 days.
Twelve New Drug Candidates Designed and Synthesized with Almost No Human Hands in the Loop
A collaboration between ETH Zurich and Goethe University Frankfurt asked a pointed question: could a generative AI model design molecules, and a microfluidics platform synthesize them, in a closed loop with minimal human interference? The target was liver X receptor (LXR) agonists, relevant to cholesterol metabolism and inflammation. Of 41 AI-designed molecules submitted for synthesis, 25 were successfully made on-chip, and 12 were confirmed as potent, previously unknown LXR agonists, some with entirely novel molecular scaffolds.
A benchtop microfluidics synthesis platform built to execute only the reaction chemistries the AI model was allowed to design around, meaning synthesizability was baked into the design step, not discovered as a bottleneck afterward. That tight coupling between design constraints and what a machine can actually build is the core promise of an automated DMTA cycle.
Turning a 16% Success Rate into 100%, Without Increasing Headcount
Compound library synthesis has a dirty secret: even experienced medicinal chemists routinely see success rates of 50-80% when producing a diversity library, and a single fixed reaction condition can perform far worse. A team from AstraZeneca and the University of Leeds tested this directly, running 900 individual amide-coupling reactions across 25 substrate combinations, four coupling agents, and nine physical conditions, completed in 192 hours.
What made it possible was a stopped-flow reactor integrated into a high-throughput liquid handling platform, run against a systematic design-of-experiments matrix rather than a single guessed condition. Screening all combinations pushed the library success rate from 16% under a single fixed condition to 100%, while using roughly 90% less reagent than a fully continuous flow approach. The resulting dataset also trained a model that could predict successful conditions with 92% accuracy on unseen chemistry.
Although HighRes did not directly contribute to these first few advancements, we're excited to see the larger automation community helping to drive progress.
A Screen That Found the First Activator No One Had Ever Reported
Researchers at AstraZeneca and the MRC Laboratory of Molecular Biology were after small molecules that could switch cytoplasmic dynein, the motor protein that hauls cargo like endosomes, mitochondria, and mRNAs along microtubules, on or off inside living cells. Inhibitors of dynein had been reported before, but no one had ever described a specific activator, despite boosting dynein-based transport being a leading hypothesis for treating age-related neurodegenerative disease, where slowed axonal transport is an early warning sign. The team built a single assay sensitive enough to catch both directions of activity at once, ran it against the full AstraZeneca compound collection, and came away with over 2,500 confirmed hits split across both directions.
Two CellInsight CX5 high-content imagers were coordinated through a Cart-to-Cart robotic system, continuously fed from a HighRes SteriStore incubator, to run more than 500,000 compounds through a live-cell trafficking assay in a single campaign. Getting the biology to cooperate took as much engineering as the imaging did: tuning the chemical inducer to a precisely timed, sub-maximal dose was what let one screen catch both agonists and antagonists, instead of needing two separate campaigns to do it. The system held a 0.09% plate failure rate across 30 runs, reliable enough to run unattended out of hours, which is what made the throughput possible in the first place. Of the hits, 85% of inhibitors and 72% of activators held up on confirmation testing, including activator chemotypes with no precedent anywhere in the prior literature.
Accelerating Science, Advancing Humanity™
The common thread isn't any single instrument; it's that liquid handling throughput, automated imaging, and DMTA-cycle discipline are now prerequisites for asking certain scientific questions at all. Every one of the results above traces back to a lab that could run more experiments, more precisely, than a person alone ever could, and that's the same conviction behind HighRes' mission of accelerating science and advancing humanity: the infrastructure in the background is what lets the science in the foreground happen. Explore our applications.
