GAYATRI VIDYA PARISHAD

COLLEGE FOR DEGREE AND PG COURSES (A)

(Affiliated to Andhra University |Accredited by NAAC With ‘A’ Grade| ISO 9001: 2015)
Visakhapatnam-530045
Department of Computer Applications (UG)

PROTEIN STRUCTURE PREDICTION

Project Guide: Project members:
Mrs. U. Sahiti 1. Y. Divya [2023-2402025]
Assistant Professor 2. G. Hari Priya [2023-2402030]
Department of Computer Applications (UG) 3. T. Sandhya Deepika [2023-2402044]
4. K. Devika [2023-2402057]
1/20
 11 ABSTRACT
ABSTRACT 66 ALGORITHM
ALGORITHM

22 INTRODUCTION
INTRODUCTION 77 EXAMPLE
EXAMPLE

EXISTING
EXISTING
33 SYSTEM
SYSTEM CONTENTS
CONTENTS 8 APPLICATIONS

PROPOSED
PROPOSED
44 SYSTEM
SYSTEM 98 FLOWCHART
FLOWCHART

55 COMPARISON
COMPARISON 109 CONCLUSION
CONCLUSION
2/20 PROTEIN STRUCTURE PREDICTION
 ABSTRACT
• Protein structure prediction is important because a protein’s 3D
shape determines its function, but models like AlphaFold do not
clearly show how reliable the predictions are.
• This project uses CNN(Convolutional Neural Network) to predict
protein 3D structures from amino acid sequences.
• A Random Forest module checks the structure by looking at
features like bond lengths and clashes, and calculates RMSD
(Root Mean Square Deviation) to show how accurate the model is.
• The system produces a PDB (Protein Data Bank) 3D model with
color-coded confidence scores for each part, making it easy to see
reliable and uncertain regions.
3/20 PROTEIN STRUCTURE PREDICTION
 INTRODUCTION
• The biological function of a protein is directly determined by its
3D structure.
• Traditional experimental methods to find protein structures are
expensive, time-consuming, and not scalable.
• Although deep learning models improve prediction accuracy,
many do not clearly show how reliable the predictions are.
• This project uses a CNN(Convolutional Neural Network) to
predict protein 3D structures from amino acid sequences.
• A Quality Assessment module using the Random Forest
algorithm evaluates reliability and provides per-residue
confidence scores.
4/20 PROTEIN STRUCTURE PREDICTION
 What Is Protein Structure?
3D Structure of a Protein

•Proteins are made of amino acids linked

together in a chain. This chain does not
stay straight - it folds into a 3D shape.

•This folded shape decides how a protein:

✓ works
✓ interacts
✓ remains stable

•The image shows one complete protein in its natural folded form.
5/20 PROTEIN STRUCTURE PREDICTION
 Main Structural Elements of a Protein:

α-Helix:
• Spiral-shaped and stable
• Usually predicted with high confidence

β-Sheet:
• Flat, sheet-like structure
• Provides strength to the protein

Loops:
• Flexible connecting regions
• More difficult to predict accurately
6/20 PROTEIN STRUCTURE PREDICTION
 Protein Structure Types -

1. Primary Structure:
•Linear sequence of amino acids
•Determines how the protein folds
•Not visible in 3D images

2. Secondary Structure:
•Local folding patterns
•Forms α-helices and β-sheets
•Stabilized by hydrogen bonds
•Visible in protein images
7/20 PROTEIN STRUCTURE PREDICTION
 3. Tertiary Structure:
•Complete 3D folded shape of a single
protein
•Formed by interactions between
secondary structures
•Determines protein function

4. Quaternary Structure:
•Association of multiple protein
chains
•Not present in all proteins
•Not visible in single-chain structures
8/20 PROTEIN STRUCTURE PREDICTION
 EXISTING SYSTEM
❑ Current protein structure prediction methods use deep
learning models to predict the 3D structure from amino acid
sequences.

❑ They usually provide only one final structure without

showing how confident the prediction is.

❑ Quality or confidence checking is done separately, making it

difficult to trust or analyze specific regions of the predicted
protein structure.
9/20 PROTEIN STRUCTURE PREDICTION
 PROPOSED SYSTEM
❑ The system predicts a protein’s 3D structure from its amino
acid sequence using CNN.

❑ It checks the structure’s quality and uses a Random Forest

model to estimate reliability based on features like bond
errors and atomic clashes.

❑ The final 3D model shows confidence scores for each part,

with color codes highlighting reliable and uncertain regions.
10/20 PROTEIN STRUCTURE PREDICTION
 COMPARISON TABLE
ASPECT EXISTING SYSTEM PROPOSED SYSTEM
Structure Prediction Uses deep learning to predict the 3D Uses CNN to better learn sequence
structure of proteins patterns and predict structure

Confidence Information Does not tell how reliable the Provides confidence scores for
prediction is each residue (0–100%)
Quality Assessment Quality checking is missing or done Includes an integrated Quality
separately Assessment (QA) module using
Random Forest
Error Detection Hard to identify weak or incorrect Clearly highlights uncertain loops
regions and low-confidence regions
Interpretability Results are hard to trust and analyze Easy to understand with color-
coded confidence visualization
Output Only a 3D protein structure 3D PDB model + confidence
report
11/20 PROTEIN STRUCTURE PREDICTION
 ALGORITHM
STEP-1: Input the Protein Sequence
Read the amino acid sequence provided by the user.

STEP-2: Extract Sequence Features

Generate relevant sequence and physicochemical features
required for structure prediction.

STEP-3: Predict Structural Patterns

Use a CNN model to learn sequence-to-structure
relationships and predict secondary structural information.
12/20 PROTEIN STRUCTURE PREDICTION
 STEP-4: Generate 3D Protein Structure
Convert the predicted structural information into a 3D
protein model in PDB format.

STEP-5: Analyze Structural Quality

Extract physical and geometric features such as atomic
clashes, bond length deviations, and hydrophobic
packing from the predicted model.

STEP-6: Estimate Prediction Reliability

Apply a Random Forest regression model to estimate
RMSD and evaluate the quality of the predicted structure.
13/20 PROTEIN STRUCTURE PREDICTION
 STEP-7: Compute Confidence Scores
Assign per-residue confidence scores (0–100%) based
on the quality assessment results.

STEP-8: Visualize Final Output

Produce a color-coded 3D visualization highlighting
high-confidence and low-confidence regions.

14/20 PROTEIN STRUCTURE PREDICTION

 EXAMPLE
•Input: Amino acid sequence (FASTA format)
Example Protein:
MKTAYIAKQRQISFVKSHFSRQDILDLWIYHTQGYFP
•Process:
• Sequence features are extracted from the input protein
• Structure is predicted using a CNN model
• A 3D protein structure is generated in PDB format

15/20
 Quality Assessment & Confidence Visualization-
QA Module: Random Forest checks structural features:
Atomic clashes
Bond lengths
Hydrophobic packing
Output: Per-residue confidence scores (0–100%)

Visualization:
Color-coded 3D model
High confidence
Medium confidence
Low confidence
16/20
 APPLICATIONS
• Drug Discovery and Drug Design
Helps in designing new drugs by understanding protein–drug interactions.

• Disease Analysis
Useful for studying disease-related proteins and identifying mutations.

• Medical and Bioinformatics Research

Supports research in understanding protein functions and biological
processes.

• Vaccine and Therapeutic Development

Assists in identifying protein targets for vaccine and therapeutic design.
17/20 PROTEIN STRUCTURE PREDICTION
 FLOWCHART
Protein Amino
START Sequence Feature
Acid Sequence
Extraction
Input

Predicted
Physical Feature Structure Prediction
Secondary & 3D
Analysis Model (CNN)
Structure

Quality
Confidence Final 3D
Assessment STOP
Estimation PDB Model
Model
18/20 PROTEIN STRUCTURE PREDICTION
 CONCLUSION
• This project presents a system that predicts protein 3D
structures from amino acid sequences using CNN.

• It evaluates the reliability of the predicted structures using a

Random Forest quality assessment module.

• Per-residue confidence visualization is included to highlight

reliable and uncertain regions in the protein model.

• This approach improves trust and usability, making protein

structure prediction more transparent and useful for
19/20 bioinformatics research. PROTEIN STRUCTURE PREDICTION
 THANK YOU

20/20 PROTEIN STRUCTURE PREDICTION