Philip A. Romero

Glioblastoma model mouse rescue from mortality by brain-targeted catalytically-enhanced angiotensin-converting enzyme 2 mutant-antibody fusion protein

The paper shows that combining ACE2dm-s1 with a STING agonist improves survival in GBM model mice, unlike losartan alone, and suggests ACE2dm-s1 modulates the tumor microenvironment.

Experimental reasoning in large language models

This study demonstrates that large language models can design, execute, and revise experiments autonomously using a robotic system.

Megascale recombination generates millions of catalytically competent PETases

This study shows that large-scale recombination can generate millions of catalytically competent PETases compatible with high-throughput screening.

Integrating complementary biological information for multi-objective enzyme engineering

This study develops a machine learning framework to engineer enzymes by integrating sparse data with evolutionary and structural information, achieving improvements in catalytic performance, yield, and stability.

Functional alignment of protein language models via reinforcement learning

This paper introduces RLXF, a framework that aligns protein language models with experimental functional goals, improving the generation of high-functioning protein variants.

Multiobjective learning and design of bacteriophage specificity

This paper uses deep learning to design bacteriophage receptor binding proteins with multiobjective specificity, showing high tunability and success rates in targeting diverse strains.

Learning protein function through autonomous experimental interaction

This paper demonstrates an AI system that autonomously interacts with biological systems to discover new protein functions through continuous experimental feedback.

Benchmarking and behavioral characterization of LLM agents for protein design

This paper benchmarks LLM agents in protein design tasks, finding they outperform hardcoded pipelines but underperform expert practice due to limited evaluation depth.

Machine learning-guided olivetolic acid cyclase engineering enables tailored cannabinoid biosynthesis in yeast

This paper uses machine learning to engineer olivetolic acid cyclase in yeast, enhancing cannabinoid production and selectivity.

Scalable and cost-efficient custom gene library assembly from oligopools

The paper introduces OMEGA, a low-cost method for assembling large gene libraries from oligopools, achieving high sequence recovery and uniformity.

Explicit representation of germline and non-germline residues improves antibody language modeling

PRISM improves antibody language modeling by distinguishing germline and non-germline residues, achieving better binding affinity correlation and controllable generation.

AlphaFast: High-throughput AlphaFold 3 via GPU-accelerated MSA construction

AlphaFast accelerates AlphaFold 3 by using GPU-accelerated MSA construction, achieving significant speedups and cost reductions while maintaining structural accuracy.

Democratizing Protein Language Model Training, Sharing and Collaboration

The paper introduces SaprotHub, a platform that simplifies training and sharing of protein language models, enabling non-experts to build and collaborate on customized models.

Biophysics-based protein language models for protein engineering

This paper introduces METL, a protein language model that integrates biophysical modeling with machine learning to predict protein properties and design functional variants.

Droplet microfluidic screening to engineer angiotensin-converting enzyme 2 (ACE2) catalytic activity

This paper develops a droplet microfluidic platform to screen ACE2 variants on native substrates, identifying K187T as a variant with fourfold increased catalytic efficiency.

Engineered lysins to modulate human gut microbiome communities

This study shows engineered lysins can target and remodel synthetic human gut microbiomes by altering species preferences through domain recombination.

Self-driving laboratories to autonomously navigate the protein fitness landscape

The paper introduces a self-driving laboratory platform called SAMPLE for autonomous protein engineering, which successfully engineers thermostable enzymes with improved thermal tolerance.

Neural network extrapolation to distant regions of the protein fitness landscape

This study evaluates how different neural network architectures extrapolate beyond training data in protein design, finding simpler models excel in local prediction while complex models explore broader sequence space.

Inferring protein fitness landscapes from laboratory evolution experiments

This paper develops a statistical learning framework to infer protein fitness landscapes from laboratory evolution data and applies it to DHFR evolution to understand epistasis and design future proteins.

Neural networks to learn protein sequence-function relationships from deep mutational scanning data

This paper develops a deep learning framework to predict protein function from sequence using deep mutational scanning data and demonstrates its ability to design new proteins with improved properties.

Machine learning-guided acyl-ACP reductase engineering for improved in vivo fatty alcohol production

This study uses machine learning to engineer acyl-ACP reductases for improved in vivo fatty alcohol production, achieving over twofold higher titers than natural enzymes.

Active and machine learning-based approaches to rapidly enhance microbial chemical production

This work introduces ActiveOpt, a machine learning approach that reduces the number of experiments needed to optimize microbial chemical production in Escherichia coli.

Discovery of human ACE2 variants with altered recognition by the SARS-CoV-2 spike protein

The paper identifies over 100 human ACE2 SNVs with altered SARS-CoV-2 spike protein recognition and shows distant residues influence the interaction.

Inferring protein sequence-function relationships with large-scale positive-unlabeled learning

This paper develops a positive-unlabeled learning framework to infer sequence-function relationships from DMS data and applies it to design stabilized enzymes.

Investigating the dynamics of microbial consortia in spatially structured environments

This study uses a microfluidic platform to quantify how spatial arrangement affects gene expression and growth in microbial consortia, showing distance acts as a low-pass filter and feedback mechanisms influence synchronization.

Rapid microbial interaction network inference in microfluidic droplets

The paper introduces MINI-Drop, a method to infer microbial interactions in microfluidic droplets, showing it can accurately determine pairwise and higher-order interactions.