canloc / templates /about.html
Biocoder09's picture
Update templates/about.html
d4025d9 verified
raw
history blame
2.32 kB
{% extends "base.html" %}
{% block content %}
<div class="main-wrap">
<section class="card about-flex">
<div class="about-text">
<h2>About CANLoc</h2>
<p>
CANLoc is a machine-learning system designed to predict the subcellular
localization of proteins directly from the protein sequence. It combines
transformer-based embeddings from the <b>ESM2</b> model
with an optimized <b>XGBoost</b> classifier trained on curated
protein datasets.
</p>
<h3>Performance & Evaluation</h3>
<p>
CANLoc achieves high accuracy, precision, recall, and F1-scores across all
classes. We additionally validate the model using:
</p>
<ul>
<li>Train/test split evaluation</li>
<li>10-fold stratified cross-validation</li>
<li>ROC curves for each class</li>
<li>Sensitivity and specificity analysis</li>
</ul>
<p>
These evaluations confirm that CANLoc predictions are reliable for academic
and research workflows.
</p>
<h3>Intended Use</h3>
<ul>
<li>Functional protein studies</li>
<li>Localization-oriented drug delivery strategy</li>
</ul>
<h3>Model Strengths</h3>
<ul>
<li>Fast and scalable for single or batch prediction</li>
<li>Transformer embeddings provide rich biological context</li>
<li>High accuracy with interpretable confidence metrics</li>
<li>No alignment or preprocessing required beyond the raw sequence</li>
</ul>
<h3>Limitations</h3>
<ul>
<li>Performance depends on sequence length and quality</li>
<li>Ambiguous sequences may reduce confidence</li>
<li>Designed for four major classes only</li>
</ul>
<p>
CANLoc represents a balance between modern deep learning and classical
machine learning methods, producing a system that is both
<b>reliable</b> and <b>lightweight enough to deploy</b>
in real-world web applications.
</p>
</div>
<div class="about-image">
<figure>
<img src="/static/images/cell_diagram.png"
alt="Eukaryotic cell diagram showing Subcellular Location">
</figure>
</div>
</section>
</div>
{% endblock %}