Robust machine learning models provide a comprehensive assessment (characterization) of the prospects of molecules for further development
Predicting the properties of molecules
08
Estimating the complexity of the synthesis
Synthetic complexity assessment based on different predictive models: SYBA, Complexity (SCScore)
07
АDME
Various pharmacological parameters: half-life, VDss, LogBB, LogP
06
Drug similarity parameters
The module provides valuable information about a molecule's compliance with drug similarity criteria: Lipinski's Rule of Five, Gose Filter, Oprea's Rule, PAINS filters and others
05
Metabolism
Potential biological activity of the molecules, their interaction with cytochromes and predicted organ-specificity
04
Ecological properties
Assessment of potential environmental impact: bioconcentration factor, acute toxicity to the aquatic environment (57455-2017)
03
Biological activity
Enables anticipation of the efficacy and safety of molecules in pharmaceutical and biotechnology applications
02
Toxicity
Prediction of toxicity (LD50, LDLo) for different animal species by different routes of administration
As well as general toxicity models such as reproductive toxicity, hepatotoxicity or cardiotoxicity, Ames Test, acute and chronic toxicity
01
Physicochemical properties
Solubility in water and DMSO, boiling point, melting point, saturated vapor pressure, density, viscosity
Link to publication
Syntelly models predict properties based on large sets of literature data and structural descriptors of compounds using modern machine learning methods. The prediction quality is not inferior to the most common foreign software products
How it works
If there are experimental values for the requested molecule in the database, Syntelly displays them, with a green "EXP" indicator next to the parameters. If there are no experimental data for a parameter, we display predicted values calculated by machine learning methods
Experimental
values
From physical chemistry to toxicity in a single solution
Reduction of time and costs
research
More than 80 predictable properties
Maximum coverage
Assistance in optimizing the process of searching for and developing new organic compounds
Research efficiency
Advantages
Applicability of models
For property prediction models, we display the applicability level of a particular machine learning model for a selected molecule as an indicator near each property on the card
This increases the reliability of profiling new compounds and enables informed strategic planning decisions for applications in organic synthesis, medicinal and pharmaceutical chemistry
The applicability calculation is based on the molecule's belonging to a region of chemical space and the characterization of the training sample. The indicator helps the user to evaluate how much the data predicted by the neural network can be applied to each specific molecule
Information transparency
Statistical parameters for each model are presented in a separate module on the platform in the "Statistics" section and are available to each user. To determine the accuracy of the models we use the following metrics: RMSE and ROC AUC
The area bounded by the ROC curve and the axis of the proportion of false positive classifications. The higher the AUC, the better the classifier is, with a value of 0.5 demonstrating the unsuitability of the chosen classification method (corresponding to random guessing)
ROC AUC
RMS error metric.
The smaller it is, the better
RMSE
Comparison of predicted results of the Syntelly platform with experimental values
Video tutorial