From complexity to clarity: development of CHM-FIEFP for predicting effective components in Chinese herbal formulas by using big data

Development of CHM-FIEFP

The CHM-FIEFP algorithm incorporates three attributes—term, gene, and P-value—derived from enrichment analyses through network pharmacology and bioinformatics approaches. This in silico model was used to list and rank active ingredient components that might possess pharmacological activities and biological functions equivalent to the overall effects of CHM formulas. The modeling was based on the Python programming language.

Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP) database

The TCMSP online database (https://old.tcmsp-e.com/tcmsp.php) was used to predict absorption, distribution, metabolism, and excretion characteristics, including oral bioavailability, drug similarity, and Caco-2 cell permeability. Oral bioavailability ≥30% and drug similarity ≥0.18 were considered to indicate potentially reliable pharmacological activity. The TCMSP database was also used to predict potential component-related gene targets.

Gene Expression Omnibus (GEO) database

GEO datasets (https://www.ncbi.nlm.nih.gov/geo/) for various samples were searched to obtain mRNA expression profiles of GC and non-cancerous adjacent tissue. The cutoff criteria were P-value <0.01 and |log₂FC| > 1. Overlapping differentially expressed targets among datasets were considered “GC-related” and subjected to further analysis.

Cytoscape software and plugins

Cytoscape open-source software can be used to construct, analyze, and visualize complex biological networks. We constructed an interaction network for the putative component-related gene targets and known GC-related gene targets, on the basis of data obtained from the Cytoscape (Version 3.2.1) plugin Bisogenet (Version 3.0.0). The topological properties^17–19 of each target in each interaction network were assessed according to degree centrality, calculated with the Cytoscape plugin CytoNCA.

Database for Annotation, Visualization, and Integrated Discovery (DAVID)

The gene targets with high degree centrality values were uploaded to the DAVID 6.8 (https://david.ncifcrf.gov/) to determine functional and pathway enrichment.

Cell cultures and reagents

Human GC cell lines (SGC-7901 and AGS) were acquired from the China Infrastructure of Cell Line Resources (China) and maintained in RPMI-1640 and Ham’s F-12K media with 10% (v/v) fetal bovine serum and 100 U/mL penicillin/streptomycin at 37°C and 5% CO₂. Reference standards of ferulic acid (ID: F103701-5g) and cinnamic acid (ID: C108495-5g) were purchased from Aladdin Corporation (China), and dissolved in dimethyl sulfoxide (≤ 0.5% volume of the working solution).

Verification of cellular pharmacodynamics

Cell viability and colony formation ability were examined as previously described⁹. Briefly, cells were exposed to ferulic acid or cinnamic acid at 0.025, 0.05, 0.1, 0.2, or 0.3 mg/mL for 24 or 48 h before determination of cell viability with CCK-8 assays (Dojindo, Japan). For colony formation assays, cells were cultured for 2 weeks before counting of colonies with 0.1% crystal violet. For cell motility assays, cells were cultured for 24 h; the trajectory and cumulative distance were monitored with an Operetta CLS High Content Analysis System and analyzed in Harmony software (PerkinElmer, Waltham, MA). The time-lapse imaging process used a ×20 long wide distance objective in digital phase contrast mode.

Statistical analyses

Control vs. low-dosage or high-dosage groups were analyzed with Student’s t-test in GraphPad Prism 6.0. P < 0.05 was considered statistically significant.

In silico screening of pathway and gene targets

Development of the CHM-FIEFP algorithm

The CHM-FIEFP algorithm was developed with the performance score assignment approach, which evaluates the degree of fitness between each term and target in the sub-enriched results with respect to the enriched results (BP and KEGG). Subsequently, the entropy weighting method was used to calculate the score for each component (Figure 1B, C). The specific calculation process was conducted as follows.

1. Update the enriched results and the sub-enriched results by removing the terms from BP and KEGG analysis results with P-value >0.05.

2. Calculate the term number ratio (R_i) for each sub-enriched result, according to the following formula:

   

   

   where N_i is the total number of terms derived from a specific ingredient of the CHM formula, and N_f is the total number of terms derived from the formula. is the number of terms overlapping between the ingredient and formula (annotation as shown in Figure 1).

3. Update the enriched results and the sub-enriched results again by removing discordant terms.

4. Calculate the target number ratio (r_i) for each sub-enriched result, according to the following formula:

   

   

   where n_i(T_i) is the number of targets corresponding to the term derived from the ingredient; n(T_i) is the number of targets corresponding to a specific term (the default selection is the “pop hits” list); is the number of targets overlapping between the term derived from the ingredient and term (the “pop hits” list); and n_f(T_i) is the count of targets derived from the CHM formula.

5. Calculate the contribution weights of the term number ratio (W_Ri) and target number ratio (W_ri) in the analysis of sub-enriched results with the entropy weight method. These can be determined as follows:

   1. Normalize the R_i and r_i obtained for each sub-enriched result. The R_i and r_i obtained for each sub-enriched result are assumed to form a vector as follows:

      

      

      Normalization against R_i and r_i yields and , which are determined as follows:

      

      

   2. Calculate the information entropy (E) of and as follows:

      

      

   3. Calculate the contribution weights (W) of and as follows:

      

6. The fitting score (F) for each sub-enriched result is finally obtained to quantify the degree of fit between the sub-enriched results (ingredient) and the enriched results (entire formula). This score is calculated as follows:

   

Use of CHM-FIEFP to identify key DSD ingredients with anti-tumor effects

After having obtained table FA and table FB (sheet FA and sheet FB in Supplementary Table S6) for DSD, as well as the sub-tables for each component (sheets 1A–6A and 1B–6B in Supplementary Table S6), we imported the data from these tables and sub-tables into the CHM-FIEFP for calculation. Among all tested components, ferulic acid (C4) had the highest fit (F) value (Table 1). Subsequent CCK-8 assays revealed dose- and time-dependent cytotoxic effects of C4 in both GC cells (Figure 3A). With 24 h exposure, the IC₅₀ of C4 was 0.127 ± 0.005 and 0.192 ± 0.001 mg/mL in SGC-7901 cells and AGS cells, respectively. The growth inhibitory effect of C4 was time-dependent within 72 h (Figure 3B). A dose-dependent decrease in clonal formation ability was observed when cells were exposed to C4 for 48 h (Figure 3C). C4 exposure also inhibited cell motility (Figure 3D). These findings collectively demonstrated the inhibitory effects of C4 on the growth and migratory ability of GC cells; these effects were comparable to those observed with DSD⁹. Importantly, these phenotypic changes closely resembled those induced by DSD treatment but were scarcely evident in cells treated with cinnamic acid (C6), which exhibited the lowest fit (F) value among all tested components (data not shown), thereby suggesting a crucial role of C4 in mediating the anti-GC effect of the formula by modulating similar biological functions and cell signaling pathways to those modulated by DSD.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Ferulic acid suppresses the growth of gastric cancer cells. (A) Viability of gastric cancer cells, assessed after 24 or 48 h of exposure to ferulic acid at 0.025–0.3 mg/mL. (B) Growth of SGC-7901 and AGS cells, evaluated after 24, 48, or 72 h exposure to ferulic acid at 0.049 and 0.086 mg/mL. (C) Colony formation, evaluated after 48 h exposure to ferulic acid. (D) Accumulated distance of gastric cancer cells, determined after 24 h exposure to ferulic acid with an Operetta CLS High Content Analysis System. **P < 0.01; ****P < 0.0001.