Sotaro Fuchigami

REV1 is a eukaryotic error-prone DNA polymerase belonging to the Y-family, with a central role in translesion DNA synthesis (TLS) to continue DNA replication even in the presence of DNA damage in the template strand. TLS is stimulated by monoubiquitination of proliferating cell nuclear antigen (PCNA), a toroidal-shaped protein functioning as a scaffold for DNA polymerases and repair enzymes. Mammals possess four types of Y-family DNA polymerases: Pol η, Pol κ, Pol ι and REV1. Among those, Pol η, Pol κ and Pol ι interact with PCNA through PCNA-binding motifs, low-affinity variants of PCNA-interacting protein box (PIP-box). To date, several studies have reported that REV1 interacts with PCNA, but identified PCNA-binding regions are inconsistent; therefore, a structural basis for interaction between REV1 and PCNA also remains unclear. Here, we identified a signature sequence conserved within vertebrates REV1 responsible for PCNA-binding. Furthermore, we unveiled a mechanism underlying the physical interaction between the PCNA-binding motif of human REV1 and PCNA by X-ray crystallography, thus revealing that REV1 binds to PCNA through a PIP-box variant located in the C-terminal side of the little finger domain. Our study provides a convincing answer for a long-standing controversy regarding the physical interaction between REV1 and PCNA.

AFM画像の歪みの原因となる探針形状を，ノイズに強い微分可能なアルゴリズムで推定するデータ解析手法を開発した．推定された探針形状で歪みを取り除くことにより，従来法より高精度な画像鮮明化を実現することができた．本手法は，他の構造データとの定量的比較を可能にする解析ツールとして期待される．

Representation learning (RL) is a universal technique for deriving low-dimensional disentangled representations from high-dimensional observations, aiding in a multitude of downstream tasks. RL has been extensively applied to various data types, including images and natural language. Here, we analyze molecular dynamics (MD) simulation data of biomolecules in terms of RL. Currently, state-of-the-art RL techniques, mainly motivated by the variational principle, try to capture slow motions in the representation (latent) space. Here, we propose two methods based on an alternative perspective on the disentanglement in the latent space. By disentanglement, we here mean the separation of underlying factors in the simulation data, aiding in detecting physically important coordinates for conformational transitions. The proposed methods introduce a simple prior that imposes temporal constraints in the latent space, serving as a regularization term to facilitate the capture of disentangled representations of dynamics. Comparison with other methods via the analysis of MD simulation trajectories for alanine dipeptide and chignolin validates that the proposed methods construct Markov state models (MSMs) whose implied time scales are comparable to those of the state-of-the-art methods. Using a measure based on total variation, we quantitatively evaluated that the proposed methods successfully disentangle physically important coordinates, aiding the interpretation of folding/unfolding transitions of chignolin. Overall, our methods provide good representations of complex biomolecular dynamics for downstream tasks, allowing for better interpretations of the conformational transitions.

Representation learning (RL) is a universal technique for deriving low-dimensional disentangled representations from high-dimensional observations, aiding a multitude of downstream tasks. RL has been extensively applied to various data types, including images and natural language. Here, we analyze molecular dynamics (MD) simulation data of biomolecules in terms of RL to obtain disentangled representations related to their conformational transitions. Currently, state-of-the-art RL techniques, which are mainly motivated by the variational principle, try to capture slow motions in the representation (latent) space. Here, we propose two methods based on alternative perspective on the disentanglement in the representation space. The methods introduce a simple prior that imposes temporal constraints in the representation space, serving as a regularization term to facilitate capturing disentangled representations of dynamics. The introduction of this simple prior aids in characterizing the conformational transitions of proteins. Indeed, comparison with other methods via the analysis of MD simulation trajectories for alanine dipeptide and chignolin validates that the proposed methods construct Markov state models (MSMs) whose implied time scales are comparable to state-of-the-art methods. By coarse-graining MSMs, we further show the methods aid to detect physically important interactions for conformational transitions. Overall, our methods provide good representations of complex biomolecular dynamics for downstream tasks, allowing for better interpretations of conformational transitions.

High-speed (HS) atomic force microscopy (AFM) can be used to observe structural dynamics of biomolecules under near-physiological conditions. In the AFM measurement, the probe tip scans an area of interest and acquires height data pixel by pixel so that the obtained AFM image contains a measurement time difference. In this study, to integrate molecular dynamics simulations with asynchronous HS-AFM movie data, we developed a particle smoother (PS) method for Bayesian data assimilation, one of the machine learning approaches, by extending the previous particle filter method. With a twin experiment with an asynchronous pseudo HS-AFM movie of a nucleosome, we found that the PS method with the pixel-by-pixel data acquisition reproduced the dynamic behavior of a nucleosome better than the previous particle filter method that ignored the data asynchronicity. We examined several frequencies of particle resampling in the PS method, and found that resampling once per one frame was optimal for reproducing the dynamic behavior. Thus, we found that the PS method with an appropriate resampling frequency is a powerful method for estimating the dynamic behavior of a target molecule from HS-AFM data with low spatiotemporal resolution.

High-speed atomic force microscopy (HS-AFM) is a powerful technique to image the structural dynamics of biomolecules. We can obtain atomic-resolution structural information from the measured AFM image by superimposing a structural model on the image. We previously developed a flexible fitting molecular dynamics (MD) simulation method that allows for modest conformational changes when superimposed on an AFM image. In this study, for a molecular motor, myosin V (which changes its chemical state), we examined whether the conformationally distinct state in each HS-AFM image could be inferred via flexible fitting MD simulation. We first built models of myosin V bound to the actin filament in two conformational states, the “down-up” and “down-down” states. Then, for the previously obtained HS-AFM image of myosin bound to the actin filament, we performed flexible-fitting MD simulations using the two states. By comparing the fitting results, we inferred the conformational and chemical states from the AFM image.

データ同化では，ベイズ統計学に基づいて観測データとシミュレーションを融合させることで，観測対象の詳細を推定することができる．このデータ同化手法の1つである粒子フィルタを高速原子間力顕微鏡データと分子シミュレーションに適用するためのプロトコルを開発し，観測不可能な特性が推定できるかどうかを検証した．

The atomic force microscopy (AFM) is a powerful tool for imaging structures of molecules bound on surfaces. To gain high-resolution structural information, one often superimposes structure models on the measured images. Motivated by high flexibility of biomolecules, we previously developed a flexible-fitting molecular dynamics (MD) method that allows protein structural changes upon superimposing. Since the AFM image largely depends on the AFM probe tip geometry, the fitting process requires accurate estimation of the parameters related to the tip geometry. Here, we performed a Bayesian statistical inference to estimate a tip radius of the AFM probe from a given AFM image via flexible-fitting molecular dynamics (MD) simulations. We first sampled conformations of the nucleosome that fit well the reference AFM image by the flexible-fitting with various tip radii. We then estimated an optimal tip parameter by maximizing the conditional probability density of the AFM image produced from the fitted structure.

High-speed atomic force microscopy (HS-AFM) can be used to observe the structural dynamics of biomolecules at the single-molecule level in real time under near-physiological conditions; however, its spatiotemporal resolution is limited. Complementarily, molecular dynamics (MD) simulations have higher spatiotemporal resolutions, albeit with some artifacts. Here, to integrate HS-AFM data and coarse-grained molecular dynamics (CG-MD) simulations, we develop a particle filter method that implements a sequential Bayesian data assimilation approach. We test the method in a twin experiment. First, we generate a reference HS-AFM movie from the CG-MD trajectory of a test molecule, a nucleosome; this serves as the "experimental measurement." Then, we perform a particle filter simulation with 512 particles, which captures the large-scale nucleosome structural dynamics compatible with the AFM movie. Comparing particle filter simulations with 8–8192 particles, we find that using greater numbers of particles consistently increases the likelihood of the whole AFM movie. By comparing the likelihoods for different ionic concentrations and timescale mappings, we find that the "true" concentration and timescale mapping can be inferred as the largest likelihood of the whole AFM movie but not that of each AFM image. The particle filter method provides a general approach for integrating HS-AFM data with MD simulations.