Representations of lipid nanoparticles using large language models for transfection efficiency prediction.

MOTIVATION: Lipid nanoparticles (LNPs) are the most widely used vehicles for mRNA vaccine delivery. The structure of the lipids composing the LNPs can have a major impact on the effectiveness of the mRNA payload. Several properties should be optimized to improve delivery and expression including biodegradability, synthetic accessibility, and transfection efficiency. RESULTS: To optimize LNPs, we developed and tested models that enable the virtual screening of LNPs with high transfection efficiency. Our best method uses the lipid Simplified Molecular-Input Line-Entry System (SMILES) as inputs to a large language model. Large language model-generated embeddings are then used by a downstream gradient-boosting classifier. As we show, our method can more accurately predict lipid properties, which could lead to higher efficiency and reduced experimental time and costs. AVAILABILITY AND IMPLEMENTATION: Code and data links available at: https://github.com/Sanofi-Public/LipoBART.

Multitarget Approach to Drug Candidates against Alzheimer's Disease Related to AChE, SERT, BACE1 and GSK3ß Protein Targets.

Alzheimer's disease is a neurodegenerative condition for which currently there are no drugs that can cure its devastating impact on human brain function. Although there are therapeutics that are being used in contemporary medicine for treatment against Alzheimer's disease, new and more effective drugs are in great demand. In this work, we proposed three potential drug candidates which may act as multifunctional compounds simultaneously toward AChE, SERT, BACE1 and GSK3ß protein targets. These candidates were discovered by using state-of-the-art methods as molecular calculations (molecular docking and molecular dynamics), artificial neural networks and multilinear regression models. These methods were used for virtual screening of the publicly available library containing more than twenty thousand compounds. The experimental testing enabled us to confirm a multitarget drug candidate active at low micromolar concentrations against two targets, e.g., AChE and BACE1.

Identification of Natural Compounds against Neurodegenerative Diseases Using In Silico Techniques.

The aim of this study was to identify new potentially active compounds for three protein targets, tropomyosin receptor kinase A (TrkA), N-methyl-d-aspartate (NMDA) receptor, and leucine-rich repeat kinase 2 (LRRK2), that are related to various neurodegenerative diseases such as Alzheimer's, Parkinson's, and neuropathic pain. We used a combination of machine learning methods including artificial neural networks and advanced multilinear techniques to develop quantitative structure⁻activity relationship (QSAR) models for all target proteins. The models were applied to screen more than 13,000 natural compounds from a public database to identify active molecules. The best candidate compounds were further confirmed by docking analysis and molecular dynamics simulations using the crystal structures of the proteins. Several compounds with novel scaffolds were predicted that could be used as the basis for development of novel drug inhibitors related to each target.

Synthesis and bioassay of improved mosquito repellents predicted from chemical structure.

Mosquito repellency data on acylpiperidines derived from the U.S. Department of Agriculture archives were modeled by using molecular descriptors calculated by CODESSA PRO software. An artificial neural network model was developed for the correlation of these archival results and used to predict the repellent activity of novel compounds of similar structures. A series of 34 promising N-acylpiperidine mosquito repellent candidates (4a-4q') were synthesized by reactions of acylbenzotriazoles 2a-2p with piperidines 3a-3f. Compounds (4a-4q') were screened as topically applied mosquito repellents by measuring the duration of repellency after application to cloth patches worn on the arms of human volunteers. Some compounds that were evaluated repelled mosquitoes as much as three times longer than N,N-diethyl-m-toluamide (DEET), the most widely used repellent throughout the world. The newly measured durations of repellency were used to obtain a superior correlation equation relating mosquito repellency to molecular structure.

A general treatment of solubility 4. Description and analysis of a PCA model for ostwald solubility coefficients.

Principal component analysis (PCA) of a large data matrix (153 solvents x 396 solutes) for Ostwald solubility coefficients (log L) resulted in a two-component model covering 98.6% of the variability. Analysis of the principal components exposed the structural characteristics of solutes and solvents that codify interactions which determine the behavior of a chemical in the surrounding media. The pattern revealed by PCA analysis distinguishes solutes according to the molecular size, functional groups, and electrostatic interactions, such as polarity and hydrogen-bonding donor and acceptor properties.

Novel carboxamides as potential mosquito repellents.

A model was developed using 167 carboxamide derivatives, from the United States Department of Agriculture archival database, that were tested as arthropod repellents over the past 60 yr. An artificial neural network employing CODESSA PRO descriptors was used to construct a quantitative structure-activity relationship model for prediction of novel mosquito repellents. By correlating the structure of these carboxamides with complete protection time, a measure of repellency based on duration, 34 carboxamides were predicted as candidate mosquito repellents. There were four additional compounds selected on the basis of their structural similarity to those predicted. The compounds were synthesized either by reaction of 1-acylbenzotriazoles with secondary amines or by reaction of acid chlorides with secondary amines in the presence of sodium hydride. The biological efficacy was assessed by duration of repellency on cloth at two dosages (25 and 2.5 micromol/cm2) and by the minimum effective dosage to prevent Aedes aegypti (L.) (Diptera: Culicidae) bites. One compound, (E)-N-cyclohexyl-N-ethyl-2-hexenamide, was superior to N,N-diethyl-3-methylbenzamide (deet) at both the high dosage (22 d versus 7 d for deet) and low dosage (5 d versus 2.5 d for deet). Only one of the carboxamides, hexahydro-1-(l-oxohexyl)-1H-azepine, had a minimum effective dosage that was equivalent or slightly better than that of deet (0.033 micromol/cm2 versus 0.047 micromol/cm2).

QSAR modeling of the antifungal activity against Candida albicans for a diverse set of organic compounds.

The molecular structures of 83 diverse organic compounds are correlated by a quantitative structure-activity relationship (QSAR) to their minimum inhibitor concentrations (MIC expressed as log(1/MIC)), involving 6 descriptors with R(2)=0.788, F=47.140, s(2)=0.130. A novel QSAR development technique is utilized combining advantages of the two frequently applied methods. The topological, electronic, geometrical, and hybrid type descriptors for the compounds were calculated by CODESSA PRO software.

QSPR modeling of flash points: an update.

Quantitative structure-property relationship (QSPR) models for the flash points of 758 organic compounds are developed using geometrical, topological, quantum mechanical and electronic descriptors calculated by CODESSA PRO software. Multilinear regression models link the structures to their reported flash point values. We also report a nonlinear model based on an artificial neural network. The results are discussed in the light of the main factors that influence the property under investigation and its modeling.

Skin permeation rate as a function of chemical structure.

Multilinear and nonlinear QSAR models were built for the skin permeation rate (Log K(p)) of a set of 143 diverse compounds. Satisfactory models were obtained by three approaches applied: (i) CODESSA PRO, (ii) Neural Network modeling using large pools of theoretical molecular descriptors, and (iii) ISIDA modeling based on fragment descriptors. The predictive abilities of the models were assessed by internal and external validations. The descriptors involved in the equations are discussed from the physicochemical point of view to illuminate the factors that influence skin permeation.

Have artificial neural networks met expectations in drug discovery as implemented in QSAR framework?

INTRODUCTION: Artificial neural networks (ANNs) are highly adaptive nonlinear optimization algorithms that have been applied in many diverse scientific endeavors, ranging from economics, engineering, physics, and chemistry to medical science. Notably, in the past two decades, ANNs have been used widely in the process of drug discovery. AREAS COVERED: In this review, the authors discuss advantages and disadvantages of ANNs in drug discovery as incorporated into the quantitative structure-activity relationships (QSAR) framework. Furthermore, the authors examine the recent studies, which span over a broad area with various diseases in drug discovery. In addition, the authors attempt to answer the question about the expectations of the ANNs in drug discovery and discuss the trends in this field. EXPERT OPINION: The old pitfalls of overtraining and interpretability are still present with ANNs. However, despite these pitfalls, the authors believe that ANNs have likely met many of the expectations of researchers and are still considered as excellent tools for nonlinear data modeling in QSAR. It is likely that ANNs will continue to be used in drug development in the future.

Indole-like Trk receptor antagonists.

The virtual screening for new scaffolds for TrkA receptor antagonists resulted in potential low molecular weight drug candidates for the treatment of neuropathic pain and cancer. In particular, the compound (Z)-3-((5-methoxy-1H-indol-3-yl)methylene)-2-oxindole and its derivatives were assessed for their inhibitory activity against Trk receptors. The IC50 values were computationally predicted in combination of molecular and fragment-based QSAR. Thereafter, based on the structure-activity relationships (SAR), a series of new compounds were designed and synthesized. Among the final selection of 13 compounds, (Z)-3-((5-methoxy-1-methyl-1H-indol-3-yl)methylene)-N-methyl-2-oxindole-5-sulfonamide showed the best TrkA inhibitory activity using both biochemical and cellular assays and (Z)-3-((5-methoxy-1-methyl-1H-indol-3-yl)methylene)-2-oxindole-5-sulfonamide was the most potent inhibitor of TrkB and TrkC.

In silico machine learning methods in drug development.

Machine learning (ML) computational methods for predicting compounds with pharmacological activity, specific pharmacodynamic and ADMET (absorption, distribution, metabolism, excretion and toxicity) properties are being increasingly applied in drug discovery and evaluation. Recently, machine learning techniques such as artificial neural networks, support vector machines and genetic programming have been explored for predicting inhibitors, antagonists, blockers, agonists, activators and substrates of proteins related to specific therapeutic targets. These methods are particularly useful for screening compound libraries of diverse chemical structures, "noisy" and high-dimensional data to complement QSAR methods, and in cases of unavailable receptor 3D structure to complement structure-based methods. A variety of studies have demonstrated the potential of machine-learning methods for predicting compounds as potential drug candidates. The present review is intended to give an overview of the strategies and current progress in using machine learning methods for drug design and the potential of the respective model development tools. We also regard a number of applications of the machine learning algorithms based on common classes of diseases.

Rational design of a series of novel amphipathic cell-penetrating peptides.

A series of novel, amphipathic cell-penetrating peptides was developed based on a combination of the model amphipathic peptide sequence and modifications based on the strategies developed for PepFect and NickFect peptides. The aim was to study the role of amphipathicity for peptide uptake and to investigate if the modifications developed for PepFect peptides could be used to improve the uptake of another class of cell-penetrating peptides. The peptides were synthesized by solid phase peptide synthesis and characterized by circular dichroism spectroscopy. Non-covalent peptide-plasmid complexes were formed by co-incubation of the peptides and plasmids in water solution. The complexes were characterized by dynamic light scattering and cellular uptake of the complexes was studied in a luciferase-based plasmid transfection assay. A quantitative structure-activity relationship (QSAR) model of cellular uptake was developed using descriptors including hydrogen bonding, peptide charge and positions of nitrogen atoms. The peptides were found to be non-toxic and could efficiently transfect cells with plasmid DNA. Cellular uptake data was correlated to QSAR predictions and the predicted biological effects obtained from the model correlated well with experimental data. The QSAR model could improve the understanding of structural requirements for cell penetration, or could potentially be used to predict more efficient cell-penetrating peptides.

Fragment-based development of HCV protease inhibitors for the treatment of hepatitis C.

A novel computational technology based on fragmentation of the chemical compounds has been used for the fast and efficient prediction of activities of prospective protease inhibitors of the hepatitis C virus. This study spans over a discovery cycle from the theoretical prediction of new HCV NS3 protease inhibitors to the first cytotoxicity experimental tests of the best candidates. The measured cytotoxicity of the compounds indicated that at least two candidates would be suitable further development of drugs.

Using artificial neural networks to predict cell-penetrating compounds.

INTRODUCTION: Membrane-cell penetration is a key property for drug candidates, particularly those related to CNS and gastrointestinal diseases. The ability to know whether a drug or compound has the ability to perform this complex characteristic in advance would save time and money for pharmaceutical companies. One robust and fast solution is to use artificial neural networks (ANNs) to predict the cell penetration of the compound candidates. AREAS COVERED: The authors review the application of ANN methods for ANN modeling in the discovery of cell-penetrating drugs. The article looks at three main systems including the BBB, gastrointestinal absorption and permeation in addition to discussing a new approach for cell-penetrating peptide discovery. This review provides the reader with an overview of the ANN methods and applications for the broader audience interested in prediction of cell penetration of drugs. EXPERT OPINION: ANNs can be successfully applied to the prediction of cell-penetrating drugs. Researchers have a broad field of applications for the use of quantitative structure-activity relationship neural networks in drug discovery and development, and can use these areas to further investigate this important pharmaceutical topic.

Prediction of Cell-Penetrating Peptides Using Artificial Neural Networks.

An investigation of cell-penetrating peptides (CPPs) by using combination of Artificial Neural Networks (ANN) and Principle Component Analysis (PCA) revealed that the penetration capability (penetrating/non-penetrating) of 101 examined peptides can be predicted with accuracy of 80%-100%. The inputs of the ANN are the main characteristics classifying the penetration. These molecular characteristics (descriptors) were calculated for each peptide and they provide bio-chemical insights for the criteria of penetration. Deeper analysis of the PCA results also showed clear clusterization of the peptides according to their molecular features.

Legitimate utilization of large descriptor pools for QSPR/QSAR models.

The use of large descriptor pools in multilinear QSAR/QSPR approaches has recently been increasingly criticized for their sensitivity to "chance correlations". Statistical experiments substituting "real descriptor" pools by random numbers were stated to demonstrate such sensitivity. While contributing positively to the improvement of the QSAR/QSPR methodology, these approaches claim complete interchangeability between the molecular descriptors used in QSAR/QSPR models and random numbers. Here, we demonstrate that when used correctly the large molecular descriptor pools are (i) not comparable with random numbers and (ii) can give very helpful QSPR conclusions.

Novel computational models for predicting dopamine interactions.

Dopamine is a crucial neurotransmitter responsible for functioning and maintenance of the nervous system. Dopamine has also been implicated in a number of diseases including schizophrenia, Parkinson's disease and drug addiction. Dopamine agonists are used in early Parkinson's disease treatment. Dopamine antagonists suppress schizophrenia. Therefore, molecules modulating dopamine receptors activity are vastly important for understanding the nervous system functioning and for the treatment of neurological diseases. In this study we describe novel computational models that efficiently predict binding affinity of the existing small molecule dopamine analogs to dopamine receptor. The model provides the set of molecular descriptors that can be used for the development of new small molecule dopamine agonists.

QSPR modeling of hyperpolarizabilities.

The polarizabilities and the first and second hyperpolarizabilities of 219 conjugated organic compounds are modeled by QSPR (quantitative structure activity relationship) based on a large pool of constitutional, topological, electronic and quantum chemical descriptors calculated by CODESSA Pro (comprehensive descriptors for structural and statistical analysis) derived solely from molecular structure. Multilinear models were developed using the BMLR (best multilinear regression) algorithm to relate the experimental (hyper)polarizabilities to their predicted values. The regression equations include AM1 (Austin model 1) calculated (hyper)polarizabilities together with the size, electrostatic and quantum chemical descriptors to compensate for the imprecision of the AM1 computational method. The results emphasize the main factors that influence (hyper)polarizability. All models were validated by the "leave-one-out" method and internal validations that confirmed the stability and good predictive ability.