A New Fingerprint and Graph Hybrid Neural Network for Predicting Molecular Properties.

Machine learning plays a role in accelerating drug discovery, and the design of effective machine learning models is crucial for accurately predicting molecular properties. Characterizing molecules typically involves the use of molecular fingerprints and molecular graphs. These are input into a multilayer perceptron (MLP) and variants of graph neural networks, such as graph attention networks (GATs). Due to the diverse types and large dimension of fingerprints, models may contain many features that are relatively irrelevant or redundant; meanwhile, although the GAT excels in handling heterogeneous graph tasks, it lacks the ability to extract collaborative information from neighboring nodes, which is crucial in scenarios where it cannot capture the joint influence of adjacent groups on atoms. To overcome these challenges, we introduce a hybrid model, combining improved GAT and MLP. In GAT, the recurrent neural network is employed to capture collaborative information. To address the dimensionality issue, we propose a feature selection algorithm, which is based on the principle of maximizing relevance while minimizing redundancy. Through experiments on 13 public data sets and 14 breast cell lines, our model demonstrates superior performance compared to state-of-the-art deep learning and traditional machine learning algorithms. Additionally, a series of ablation experiments were conducted to demonstrate the advantages of our improved version, as well as its antinoise capability and interpretability. These results indicate that our model holds promising prospects for practical applications.

Unexpectedly resisting protein adsorption on self-assembled monolayers terminated with two hydrophilic hydroxyl groups.

OH-terminated self-assembled monolayers, as protein-resistant surfaces, have significant potential in biocompatible implant devices, which can avoid or reduce adverse reactions caused by protein adhesion to biomaterial surfaces, such as thrombosis, immune response and inflammation. Here, molecular dynamics simulations were performed to evaluate the degree of protein adsorption on the self-assembled monolayer terminated with two hydrophilic OH groups ((OH)2-SAM) at packing densities (Σ) of 4.5 nm-2 and 6.5 nm-2, respectively. The results show that the structure of the (OH)2-SAM itself, i.e., a nearly perfect hexagonal-ice-like hydrogen bond structure in the OH matrix of the (OH)2-SAM at Σ = 4.5 nm-2 sharply reduces the number of hydrogen bonds (i.e., 0.7 ± 0.27) formed between the hydrophobic (OH)2-SAM surface and protein. While for Σ = 6.5 nm-2, the hydrophilic (OH)2-SAM surface can provide more hydrogen bonding sites to form hydrogen bonds (i.e., 6.2 ± 1.07) with protein. The number of hydrogen bonds formed between the (OH)2-SAM and protein at Σ = 6.5 nm-2 is ∼8 times higher than that at Σ = 4.5 nm-2, reflecting the excellent resistance to protein adsorption exhibited by the structure of the (OH)2-SAM itself at Σ = 4.5 nm-2. Compared with a traditional physical barrier effect formed by a large number of hydrogen bonds between the (OH)2-SAM and water at Σ = 6.5 nm-2, the structure of the (OH)2-SAM itself at Σ = 4.5 nm-2 proposed in this study significantly improves the performance of the (OH)2-SAM resistance to protein adsorption, which provides new insights into the mechanism of resistance to protein adsorption on the (OH)2-SAM.

Radial distribution function of semiflexible oligomers with stretching flexibility.

The radial distribution of the end-to-end distance Ree is crucial for quantifying the global size and flexibility of a linear polymer. For semiflexible polymers, several analytical formulas have been derived for the radial distribution of Ree ignoring the stretching flexibility. However, for semiflexible oligomers, such as DNA or RNA, the stretching flexibility can be rather pronounced and can significantly affect the radial distribution of Ree. In this study, we obtained an extended formula that includes the stretch modulus to describe the distribution of Ree for semiflexible oligomers on the basis of previous formulas for semiflexible polymers without stretching flexibility. The extended formula was validated by extensive Monte Carlo simulations over wide ranges of the stretch modulus and persistence length, as well as all-atom molecular dynamics simulations of short DNAs and RNAs. Additionally, our analyses showed that the effect of stretching flexibility on the distribution of Ree becomes negligible for DNAs longer than ∼130 base pairs and RNAs longer than ∼240 base pairs.

Divalent Ion-Mediated DNA-DNA Interactions: A Comparative Study of Triplex and Duplex.

Ion-mediated interaction between DNAs is essential for DNA condensation, and it is generally believed that monovalent and nonspecifically binding divalent cations cannot induce the aggregation of double-stranded (ds) DNAs. Interestingly, recent experiments found that alkaline earth metal ions such as Mg2+ can induce the aggregation of triple-stranded (ts) DNAs, although there is still a lack of deep understanding of the surprising findings at the microscopic level. In this work, we employed all-atom dynamic simulations to directly calculate the potentials of mean force (PMFs) between tsDNAs, between dsDNAs, and between tsDNA and dsDNA in Mg2+ solutions. Our calculations show that the PMF between tsDNAs is apparently attractive and becomes more strongly attractive at higher [Mg2+], although the PMF between dsDNAs cannot become apparently attractive even at high [Mg2+]. Our analyses show that Mg2+ internally binds into grooves and externally binds to phosphate groups for both tsDNA and dsDNA, whereas the external binding of Mg2+ is much stronger for tsDNA. Such stronger external binding of Mg2+ for tsDNA favors more apparent ion-bridging between helices than for dsDNA. Furthermore, our analyses illustrate that bridging ions, as a special part of external binding ions, are tightly and positively coupled to ion-mediated attraction between DNAs.

Understanding the Relative Flexibility of RNA and DNA Duplexes: Stretching and Twist-Stretch Coupling.

The flexibility of double-stranded (ds) RNA and dsDNA is crucial for their biological functions. Recent experiments have shown that the flexibility of dsRNA and dsDNA can be distinctively different in the aspects of stretching and twist-stretch coupling. Although various studies have been performed to understand the flexibility of dsRNA and dsDNA, there is still a lack of deep understanding of the distinctive differences in the flexibility of dsRNA and dsDNA helices as pertains to their stretching and twist-stretch coupling. In this work, we have explored the relative flexibility in stretching and twist-stretch coupling between dsRNA and dsDNA by all-atom molecular dynamics simulations. The calculated stretch modulus and twist-stretch coupling are in good accordance with the existing experiments. Our analyses show that the differences in stretching and twist-stretch coupling between dsRNA and dsDNA helices are mainly attributed to their different (A- and B-form) helical structures. Stronger basepair inclination and slide in dsRNA is responsible for the apparently weaker stretching rigidity versus that of dsDNA, and the opposite twist-stretch coupling for dsRNA and dsDNA is also attributed to the stronger basepair inclination in dsRNA than in dsDNA. Our calculated macroscopic elastic parameters and microscopic analyses are tested and validated by different force fields for both dsRNA and dsDNA.

Multivalent ion-mediated nucleic acid helix-helix interactions: RNA versus DNA.

Ion-mediated interaction is critical to the structure and stability of nucleic acids. Recent experiments suggest that the multivalent ion-induced aggregation of double-stranded (ds) RNAs and DNAs may strongly depend on the topological nature of helices, while there is still lack of an understanding on the relevant ion-mediated interactions at atomistic level. In this work, we have directly calculated the potentials of mean force (PMF) between two dsRNAs and between two dsDNAs in Co(NH3)6 (3+) (Co-Hex) solutions by the atomistic molecular dynamics simulations. Our calculations show that at low [Co-Hex], the PMFs between B-DNAs and between A-RNAs are both (strongly) repulsive. However, at high [Co-Hex], the PMF between B-DNAs is strongly attractive, while those between A-RNAs and between A-DNAs are still (weakly) repulsive. The microscopic analyses show that for A-form helices, Co-Hex would become 'internal binding' into the deep major groove and consequently cannot form the evident ion-bridge between adjacent helices, while for B-form helices without deep grooves, Co-Hex would exhibit 'external binding' to strongly bridge adjacent helices. In addition, our further calculations show that, the PMF between A-RNAs could become strongly attractive either at very high [Co-Hex] or when the bottom of deep major groove is fixed with a layer of water.

Flexibility of short DNA helices with finite-length effect: From base pairs to tens of base pairs.

Flexibility of short DNA helices is important for the biological functions such as nucleosome formation and DNA-protein recognition. Recent experiments suggest that short DNAs of tens of base pairs (bps) may have apparently higher flexibility than those of kilo bps, while there is still the debate on such high flexibility. In the present work, we have studied the flexibility of short DNAs with finite-length of 5-50 bps by the all-atomistic molecular dynamics simulations and Monte Carlo simulations with the worm-like chain model. Our microscopic analyses reveal that short DNAs have apparently high flexibility which is attributed to the significantly strong bending and stretching flexibilities of ∼6 bps at each helix end. Correspondingly, the apparent persistence length lp of short DNAs increases gradually from ∼29 nm to ∼45 nm as DNA length increases from 10 to 50 bps, in accordance with the available experimental data. Our further analyses show that the short DNAs with excluding ∼6 bps at each helix end have the similar flexibility with those of kilo bps and can be described by the worm-like chain model with lp ∼ 50 nm.

A coarse-grained model with implicit salt for RNAs: predicting 3D structure, stability and salt effect.

To bridge the gap between the sequences and 3-dimensional (3D) structures of RNAs, some computational models have been proposed for predicting RNA 3D structures. However, the existed models seldom consider the conditions departing from the room/body temperature and high salt (1M NaCl), and thus generally hardly predict the thermodynamics and salt effect. In this study, we propose a coarse-grained model with implicit salt for RNAs to predict 3D structures, stability, and salt effect. Combined with Monte Carlo simulated annealing algorithm and a coarse-grained force field, the model folds 46 tested RNAs (≤45 nt) including pseudoknots into their native-like structures from their sequences, with an overall mean RMSD of 3.5 Å and an overall minimum RMSD of 1.9 Å from the experimental structures. For 30 RNA hairpins, the present model also gives the reliable predictions for the stability and salt effect with the mean deviation ∼ 1.0 °C of melting temperatures, as compared with the extensive experimental data. In addition, the model could provide the ensemble of possible 3D structures for a short RNA at a given temperature/salt condition.

Salt contribution to the flexibility of single-stranded nucleic acid offinite length.

Nucleic acids are negatively charged macromolecules and their structure properties are strongly coupled to metal ions in solutions. In this article, the salt effects on the flexibility of single-stranded (ss) nucleic acid chain ranging from 12 to 120 nucleotides are investigated systematically by the coarse-grained Monte Carlo simulations where the salt ions are considered explicitly and the ss chain is modeled with the virtual-bond structural model. Our calculations show that, the increase of ion concentration causes the structural collapse of ss chain and multivalent ions are much more efficient in causing such collapse, and both trivalent/small divalent ions can induce more compact state than a random relaxation state. We found that monovalent, divalent, and trivalent ions can all overcharge ss chain, and the dominating source for such overcharging changes from ion-exclusion-volume effect to ion Coulomb correlations. In addition, the predicted Na(+) and Mg(2+)-dependent persistence length l(p)'s of ss nucleic acid are in accordance with the available experimental data, and through systematic calculations, we obtained the empirical formulas for l(p) as a function of [Na(+)], [Mg(2+)] and chain length.

[Preliminary study on rDNA ITS sequencing and characteristics of Citrus grandis var. tomentosa].

OBJECTIVE: To study the difference of rDNA ITS sequences between Cirtus grandis var. tomentosa and its relatives. METHODS: The rDNA ITS gene fragment was amplified by PCR with universal primers of rDNA ITS and sequenced. RESULTS: The DNA sequences of ITS ranged from 675 to 683 bp and the 5.8S gene of all of the samples are 165 bp in size. Distance values of ITS sequence, estimated according to Kimura two-parameter models between C. grandis var. tomentosa and C. grandis var. shatin-you, C. grandis var. tomentosa and C. reticulata were 1.05 and 2.29 respectively. CONCLUSION: There are some differences in rDNA ITS sequences among C. grandis var. tomentosa and its relatives.