{% extends "base.html" %} {% block title %}Drug-Likeness Predictor โ€” MoleSight AI{% endblock %} {% block content %}

Drug-Likeness ML Predictor

Gradient Boosting Classifier ยท 5-Fold CV ยท Trained on 1,500 synthetic compounds
SMILES Input
Quick Examples
{% for name, smi in examples %} {{ name }} {% endfor %}
{% if error %}
โš  {{ error }}
{% endif %} {% if result %}
Prediction Result
{{ (result.pred.probability * 100)|round|int }}%
{{ result.pred.label }}
Confidence: {{ result.pred.confidence }}
5-Fold CV Accuracy: {{ (result.cv_accuracy * 100)|round(1) }}%
Input Features (Model View)
{% set p = result.props %}
FeatureValue
Molecular Weight{{ p.mw }} Da
LogP{{ p.logp }}
H-Bond Donors{{ p.hbd }}
H-Bond Acceptors{{ p.hba }}
TPSA{{ p.tpsa }} ร…ยฒ
Rotatable Bonds{{ p.rotbonds }}
QED Score{{ p.qed }}
Lipinski Ro5 Assessment
{% if result.lipinski.passed %} โœ“ COMPLIANT {% else %} โœ— {{ result.lipinski.violations }} VIOLATION(S) {% endif %} {{ result.lipinski.label }}
{% for rule, passed in result.lipinski.rules.items() %}
{{ 'โœ“' if passed else 'โœ—' }} {{ rule }}
{% endfor %}
{% endif %}
{% if result %}
Prediction Probability Gauge
Drug-Likeness Profile (Radar)
Feature Importance
{% else %}
๐Ÿง 
The Gradient Boosting model uses 7 molecular descriptors to predict whether a compound is drug-like. Enter a SMILES string to see the prediction.
Model Architecture

Algorithm โ€” Gradient Boosting Classifier (scikit-learn)

Training Data โ€” 1,500 synthetic compounds with realistic property distributions

Features โ€” MW, LogP, HBD, HBA, TPSA, RotBonds, QED

Evaluation โ€” 5-fold cross validation accuracy

{% endif %}
{% endblock %} {% block scripts %} {% endblock %}