Paul Kim

Mr. Paul Kim

orcid

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
18 September 2025 in bioRxiv
Verified
Preprint
Open access
JBJasper Butcher
Jasper Butcher
RKRohith Krishna
Rohith Krishna
RMRaktim Mitra
Raktim Mitra
RBRafael I. Brent
Rafael I. Brent
YLYanjing Li
Yanjing Li
DBDavid Baker
David Baker
Deep learning has accelerated protein design, but most existing methods are restricted to generating protein backbone coordinates and often neglect interactions with other biomolecules. We present RFdiffusion3 (RFD3), a diffusion model that generates protein structures in the context of ligands, nucleic acids and other non-protein constellations of atoms. Because all polymer atoms are modeled explicitly, conditioning the model on complex sets of atom-level constraints for enzyme design and other challenges is both simpler and more effective than previous approaches. RFD3 achieves improved performance compared to prior approaches on a range of in silico benchmarks with one tenth the computational cost. Finally, we demonstrate the broad applicability of RFD3 by designing and experimentally characterizing DNA binding proteins and cysteine hydrolases. The ability to rapidly generate protein structures guided by complex sets of atom-level constraints in the context of arbitrary non-protein atoms should further expand the range of functions attainable through protein design.

1 January 2023 in IUCrJ
Verified
Journal Article
Open access
PDF file
ACAnchi Cheng
Anchi Cheng
PKPaul Kim
Paul Kim
HKHuihui Kuang
Huihui Kuang
JMJoshua H. Mendez
Joshua H. Mendez
ECEugene Y. D. Chua
Eugene Y. D. Chua
ANAlex J. Noble
Alex J. Noble
Single-particle cryo-electron microscopy (cryoEM) is a swiftly growing method for understanding protein structure. With increasing demand for high-throughput, high-resolution cryoEM services comes greater demand for rapid and automated cryoEM grid and sample screening. During screening, optimal grids and sample conditions are identified for subsequent high-resolution data collection. Screening is a major bottleneck for new cryoEM projects because grids must be optimized for several factors, including grid type, grid hole size, sample concentration, buffer conditions, ice thickness and particle behavior. Even for mature projects, multiple grids are commonly screened to select a subset for high-resolution data collection. Here, machine learning and novel purpose-built image-processing and microscope-handling algorithms are incorporated into the automated data-collection software Leginon, to provide an open-source solution for fully automated high-throughput grid screening. This new version, broadly called Smart Leginon, emulates the actions of an operator in identifying areas on the grid to explore as potentially useful for data collection. Smart Leginon Autoscreen sequentially loads and examines grids from an automated specimen-exchange system to provide completely unattended grid screening across a set of grids. Comparisons between a multi-grid autoscreen session and conventional manual screening by 5 expert microscope operators are presented. On average, Autoscreen reduces operator time from ∼6 h to <10 min and provides a percentage of suitable images for evaluation comparable to the best operator. The ability of Smart Leginon to target holes that are particularly difficult to identify is analyzed. Finally, the utility of Smart Leginon is illustrated with three real-world multi-grid user screening/collection sessions, demonstrating the efficiency and flexibility of the software package. The fully automated functionality of Smart Leginon significantly reduces the burden on operator screening time, improves the throughput of screening and recovers idle microscope time, thereby improving availability of cryoEM services.

1 January 2023 in IUCrJ
Verified
Journal Article
Open access
PDF file
PKPaul Kim
Paul Kim
ANAlex J. Noble
Alex J. Noble
ACAnchi Cheng
Anchi Cheng
TBTristan Bepler
Tristan Bepler
Over the past decade, cryo-electron microscopy (cryoEM) has emerged as an important method for determining near-native, near-atomic resolution 3D structures of biological macromolecules. To meet the increasing demand for cryoEM, automated methods that improve throughput and efficiency of microscope operation are needed. Currently, the targeting algorithms provided by most data-collection software require time-consuming manual tuning of parameters for each grid, and, in some cases, operators must select targets completely manually. However, the development of fully automated targeting algorithms is non-trivial, because images often have low signal-to-noise ratios and optimal targeting strategies depend on a range of experimental parameters and macromolecule behaviors that vary between projects and collection sessions. To address this, Ptolemy provides a pipeline to automate low- and medium-magnification targeting using a suite of purpose-built computer vision and machine-learning algorithms, including mixture models, convolutional neural networks and U-Nets. Learned models in this pipeline are trained on a large set of images from real-world cryoEM data-collection sessions, labeled with locations selected by human operators. These models accurately detect and classify regions of interest in low- and medium-magnification images, and generalize to unseen sessions, as well as to images collected on different microscopes at another facility. This open-source, modular pipeline can be integrated with existing microscope control software to enable automation of cryoEM data collection and can serve as a foundation for future cryoEM automation software.

Verified
Journal Article
Open access
PDF file
PKPaul Kim
Paul Kim
RWRobin Winter
Robin Winter
DCDjork-Arné Clevert
Djork-Arné Clevert
In silico protein–ligand binding prediction is an ongoing area of research in computational chemistry and machine learning based drug discovery, as an accurate predictive model could greatly reduce the time and resources necessary for the detection and prioritization of possible drug candidates. Proteochemometric modeling (PCM) attempts to create an accurate model of the protein–ligand interaction space by combining explicit protein and ligand descriptors. This requires the creation of information-rich, uniform and computer interpretable representations of proteins and ligands. Previous studies in PCM modeling rely on pre-defined, handcrafted feature extraction methods, and many methods use protein descriptors that require alignment or are otherwise specific to a particular group of related proteins. However, recent advances in representation learning have shown that unsupervised machine learning can be used to generate embeddings that outperform complex, human-engineered representations. Several different embedding methods for proteins and molecules have been developed based on various language-modeling methods. Here, we demonstrate the utility of these unsupervised representations and compare three protein embeddings and two compound embeddings in a fair manner. We evaluate performance on various splits of a benchmark dataset, as well as on an internal dataset of protein–ligand binding activities and find that unsupervised-learned representations significantly outperform handcrafted representations.

15 October 2026 in Science
Verified
Journal Article
Closed access
JBJasper Butcher
Jasper Butcher
RKRohith Krishna
Rohith Krishna
RMRaktim Mitra
Raktim Mitra
SWSeth M. Woodbury
Seth M. Woodbury
DKDonghyo Kim
Donghyo Kim
DBDavid Baker
David Baker
The catalytic and other sophisticated functions of biomolecules arise from the placement of many atoms in three-dimensional space. Here, we describe a generative artificial intelligence method for biopolymers, RFdiffusion3 (RFD3), which employs free atom diffusion to satisfy complex sets of atomic constraints more rapidly and accurately than previous methods across several challenges from enzyme design to RNA structure design. We demonstrate experimentally the power of the RFD3 pipeline for enzyme design by designing de novo phosphotriesterases with a complex active site architecture incorporating a carbamylated lysine coordinating two metal ions and zinc proteases specific for the transthyretin protein involved in amyloid diseases; in both cases we obtain active catalysts in zero shot 96 construct design campaigns.

27 April 2026 in bioRxiv
Verified
Preprint
Open access
PDF file
ESEnisha Sehgal
Enisha Sehgal
YPYuliya Politanska
Yuliya Politanska
RMRaktim Mitra
Raktim Mitra
PKPaul Kim
Paul Kim
NRNayim Gonzalez Rodriguez
Nayim Gonzalez Rodriguez
DBDavid Baker
David Baker
De novo protein design has advanced rapidly in recent years, yet the programmable recognition of specific DNA sequences remains a longstanding challenge. Here we describe a deep learning based approach for designing sequence selective DNA binding proteins. Our method combines structure generation using RFdiffusion3 with explicit screening against off-target interactions using AlphaFold3. We test this approach by generating 96 designs for each of 15 diverse DNA targets and identify specific binders for 7 targets, representing a ~100-fold improvement in success rates over previous approaches. We further characterize the binding landscape using variant competition assays and randomized library screening, revealing robust sequence discrimination across diverse targets. Together, these results represent a significant step forward in de novo sequence specific DNA binder design.