CC-DPS Logo

This page provides a summary of how we validate the reliability of the information provided by our Chemical Compounds Deep Profiling Services (CC-DPS). Large-scale experimental data have been collected and refined for QSQN model development and reliability validation of CC-DPS information. The thermo-physicochemical, thermodynamic, transport, and pharmaceutical properties of all compounds with available experimental data have undergone rigorous validation. Our property information has consistently demonstrated an average accuracy exceeding 95%.

In cases where experimental data for a target property is not available, we analyze the property by comparing it with similar compounds for which experimental data exist. This comparison is supplemented with the information obtained from other existing methods. Our deep profiling services' information has been widely used by over 1 million researchers worldwide and has been cited in prestigious scientific journals, including NATURE, ELSEVIER, and publications from the American Chemical Society.

Over a period of 5 years, we have amassed a comprehensive collection of experimental property data, encompassing over 1.5 million data points for more than 230,000 chemical compounds. This data has been sourced from over 160,000 diverse origins, including journal articles, scientific books, patents, and existing chemical database products. Upon detecting the frequent occurrence of significant errors in the collected experimental data, we have undertaken systematic refinement to establish reliable data points. Our data refinement process involves basic analysis, statistical filtering, and similarity analysis. For an illustrative guide to these procedures, please refer to our experimental refinement example webpage, which uses the normal boiling point as a case study.

Each of the properties produced by our services has been validated with the corresponding refined experimental data points.

The parity plot on the left demonstrates the alignment between produced and experimental values, ideally clustering close to the 45-degree line. The vertical grey lines represent the ranges of the various experimental data points that were collected. On the right, the accuracy distribution chart provides a statistical overview, showing the percentage of produced values falling within specific deviation ranges from the experimental values, thereby illustrating the accuracy of our data. Over 95% of the produced values deviate by less than ±1.5% from the refined experimental data.

Value-to-value comparisons are also conducted between our produced data and the refined experimental data. The table below, for example, shows a comparison of the normal boiling points (in Kelvin) for 100 selected compounds, including the refined experimental data for each compound along with the minimum and maximum experimental values. View Comparison Table

Parity plots, accuracy distribution charts, and value-to-value comparisons for other constant properties are also available and listed below, along with their corresponding links. View Other Properties

The plot features a red line representing CC-DPS values and blue circles for the refined experimental data, illustrating the reliability of the CC-DPS values. Additional examples of reliability verification for other temperature-dependent properties are also available and listed below, along with their corresponding links. View Other Properties

For comparative analysis, CC-DPS additionally offers estimations of certain properties using various established methods. Approaches for property estimation, reviewed by Poling et al., encompass group contribution methods and QSPR (Quantitative Structure Property Relationship) techniques, which have been utilized for decades. We have included renowned methods such as those of Joback and Gani, which are widely used in a range of industrial applications, including chemical process simulation software like Aspen Plus. The table below summarizes these existing approaches that CC-DPS provides. View Table

The reliability validation of other existing methods has also been conducted. For instance, the parity plot and the accuracy distribution chart for Joback and Gani’s approaches are presented below, using the normal boiling point as an example:

As the boiling point value increases, we observed that Joback tends to underpredict, whereas Gani often overpredicts. In Joback’s case, 64.68% of the values are within ±1.5% deviation from the refined experimental data, compared to 79.19% in Gani’s case.

Generally, many of these existing approaches show limitations in reliability, particularly when applied to complex chemical compounds with numerous heavy atoms and/or multiple functional groups. The accuracy tends to decrease, likely due to the empirical nature of the formulas and parameters. For lighter compounds, however, the data from these methods can be useful as supplementary information, offering a rough estimate of the property values.

Further examples of reliability verification for these approaches are available and can be accessed below, along with their respective links. View Other Examples

CC-DPS initially validates the produced values against corresponding refined experimental data when available. These values are considered reliable if they closely align with the experimental data and/or fall within the typical experimental error range. In the absence of experimental data, expert inspection is employed, involving a comparative analysis with similar compounds that have available experimental data.

Similar compounds are automatically determined using algorithms like Tanimoto or based on the squared correlation coefficient of molecular descriptors. The CC-DPS values are then analyzed in conjunction with all available data from these similar compounds, encompassing both experimental information and the values produced by other methods.

The information offered by CC-DPS has been used by a diverse group of users worldwide, including over 1 million individuals, 2,900 universities, 1,800 companies, and 200 organizations, across various sectors of chemical applications. A detailed list is available (Note: 'Mol-Instincts' shown in the image refers to the CC-DPS database system, representing the database constructed with CC-DPS information).

The values provided by CC-DPS have been cited numerous times in high-impact scientific journals, such as NATURE, ELSEVIER, Springer, the American Chemical Society, the Royal Society of Chemistry, and Wiley. Below is a partial list of these publications. View Citation List