What Makes Chemicals Biodegradable? A Data-Driven Guide

By shrijeetverma13 · May 29, 2026

This analysis reveals which molecular properties best predict whether chemicals break down naturally in the environment. We identified the strongest…

The analysis produced two visualizations — a bar chart and a pie chart — showing the distribution of biodegradable versus non-biodegradable compounds in the dataset. The Class variable contains two categories: Ready Biodegradable (RB, Class=1) and Not Ready Biodegradable (NRB, Class=2). Both charts clearly illustrate how the compounds are split between these two groups.

The analysis identified the top features most strongly correlated with the Class outcome variable. V1 is the strongest predictor (r = -0.3961), followed closely by V27 (r = -0.3900) and V22 (r = -0.3723). Notably, all top correlated features are negatively correlated with Class, meaning higher values in these features tend to be associated with Class 1 (readily biodegradable). Two visualizations were generated: a horizontal bar chart showing the top 15 features by correlation strength, and a heatmap displaying the correlation matrix among the top 10 features plus Class.

All three features — V8, V23, and V30 — display extreme value ranges that suggest influential outliers, but to varying degrees. V23 (integer values, range 0–147) is the most skewed, with a very low median but a maximum of 147, indicating strong right-tail outliers. V30 (float, range 0–71.17) is heavily zero-inflated, meaning most values cluster at zero while a few extreme values reach ~71, making its outliers especially influential. V8 (float, range 0–60.7) shows a more moderate spread but still exhibits notable outliers given its zero floor and 60.7 maximum. Two visualizations were generated: a box plot comparing the distributions and outlier points for all three features side by side, and a bar chart showing outlier counts per feature using the IQR method.

The analysis identified feature pairs with strong multicollinearity in your dataset. Two visualizations were generated: a correlation heatmap showing the relationships among the most correlated features, and a horizontal bar chart ranking the top 15 feature pairs by absolute correlation strength. Features with r ≥ 0.8 are flagged as strong candidates for consolidation — you can retain one from each pair, average them, or apply PCA to reduce redundancy before modeling.

The analysis identified the numeric features with the largest mean differences between Class 1 and Class 2. Feature V8 stands out as the most differentiated, with an absolute mean difference of 3.897 (Class 1 mean: 38.371 vs Class 2 mean: 34.474). Two bar charts and supporting data tables were generated to visualize these differences.

All three binary features (V24, V25, V29) significantly increase biodegradability rates when their value is 1. The overall dataset biodegradability rate is 66.3%. When any of these features equals 1, compounds are far more likely to be readily biodegradable — reaching rates above 92–95%. Two bar charts and detailed data tables were generated to visualize these differences clearly.

IQR-based outlier detection was performed on 8 continuous features. The results are displayed in a box plot colored by Class (biodegradable vs non-biodegradable) and a summary table. V36 has the most outliers (81, 7.68%), followed by V31 (77, 7.30%) and V39 (65, 6.16%). V8 is the cleanest feature with only 4 outliers (0.38%). The box plots show how outlier distribution differs between the two compound classes across all 8 features.

The analysis of continuous features V1, V8, V15, V31, and V39 has been completed. A histogram plot showing the distribution of all five features was generated, along with a skewness and kurtosis comparison chart and a statistics summary table. These visualizations reveal the shape, spread, and central tendency of each feature's distribution.

A comprehensive summary statistics table was generated for all 15 float/continuous columns grouped by Class (1 and 2). For each group, mean, median, standard deviation, min, max, and IQR were computed. Two bar chart visualizations were also produced — one comparing mean values across all features by class, and another highlighting the top 5 most discriminative features. The top 5 features with the largest absolute mean difference between Class 1 and Class 2 are: V8 (3.90), V39 (0.96), V31 (0.77), V15 (0.67), and V36 (0.61).

The Pearson correlation analysis of 15 continuous molecular descriptors (V1–V39) has been completed. Two visualizations were generated: a full annotated heatmap showing all pairwise correlations across the 15 features, and a horizontal bar chart highlighting the top 10 feature pairs ranked by absolute correlation strength. The heatmap uses a Red-Blue color scale where deep red indicates strong positive correlation and deep blue indicates strong negative correlation. The bar chart clearly distinguishes positive (blue) and negative (red) correlations with exact values labeled. Multiple data tables were also produced capturing the top positively correlated pairs, top negatively correlated pairs, and any multicollinearity flags where r 0.7. These results allow you to quickly identify which molecular descriptors carry redundant information and which pairs may cause instability in regression-based QSAR models.