B.Tech. / M.Tech (Integrated) DEGREE EXAMINATION, DECEMBER 2023
21BTB102T - INTRODUCTION TO COMPUTATIONAL BIOLOGY

Detailed Solution:
Step 1: Meiosis is a specialized type of cell division that reduces the chromosome number by half.
Step 2: It involves one round of DNA replication followed by two rounds of cell division.
Step 3: As a result, a single diploid parent cell ($2n$) produces four haploid ($n$) daughter cells, which function as gametes (sperm and egg) in sexual reproduction.

Detailed Solution:
Step 1: Artificial Neural Networks are inspired by the nervous system. Artificial Immune Systems are inspired by the immune system.
Step 2: A Genetic Algorithm (GA) is a metaheuristic search algorithm used to solve optimization problems.
Step 3: It is directly inspired by Charles Darwin's theory of natural evolution, mimicking biological processes such as selection, crossover (recombination), and mutation to "evolve" better solutions over generations.

Detailed Solution:
Step 1: The peripheral nervous system is divided into two main pathways: afferent and efferent.
Step 2: The afferent pathway (sensory) carries signals towards the Central Nervous System (CNS - brain/spinal cord).
Step 3: The efferent pathway (motor) carries signals away from the CNS to the effector organs. Therefore, signals traveling from the brain to motor neurons (which then innervate muscles) constitute the efferent pathway.

Detailed Solution:
Step 1: During prophase, chromatin condenses into visible chromosomes. During metaphase, chromosomes align at the cell equator.
Step 2: During anaphase, the cohesin proteins binding the sister chromatids together break down.
Step 3: The spindle fibers contract, pulling the sister chromatids apart toward opposite poles of the cell.

Detailed Solution:
Step 1: Ribosomes themselves are physically constructed of ribosomal RNA (rRNA) and proteins.
Step 2: However, during the process of translation, it is the messenger RNA (mRNA) that temporarily attaches to the small subunit of the ribosome.
Step 3: The ribosome acts as a workbench that clamps down on the attached mRNA strand to read its codons and synthesize a protein.

Detailed Solution:
Step 1: Glycosidic bonds link sugars (carbohydrates). Peptide bonds link amino acids (proteins).
Step 2: In a nucleic acid strand (DNA or RNA), individual nucleotide monomers are linked together.
Step 3: The 3' hydroxyl (-OH) group of the sugar of one nucleotide reacts with the 5' phosphate group of the next nucleotide, forming a strong covalent phosphodiester bond. This creates the sugar-phosphate backbone.

Detailed Solution:
Step 1: The question asks to find the "odd one out" regarding the differences between DNA and RNA.
Step 2: (A) Strandedness (RNA is single, DNA is double), (C) Nitrogenous base (RNA has Uracil, DNA has Thymine), and (D) Stability (DNA is much more stable than RNA) are all valid differences.
Step 3: Both DNA and RNA possess a phosphate group attached to the 5' carbon of their respective pentose sugars, meaning both have a 5' oxygen involved in the phosphodiester linkage. Therefore, this is a similarity, making it the odd one out among the listed differences.

Detailed Solution:
Step 1: GenBank, EMBL, and DDBJ are the primary databases for nucleotide (DNA/RNA) sequences.
Step 2: UniProtKB (Universal Protein Knowledgebase) is the central hub for the collection of functional information on proteins.
Step 3: It consists of two sections: Swiss-Prot (manually annotated and reviewed protein sequences) and TrEMBL (automatically annotated and unreviewed protein sequences).

Detailed Solution:
Step 1: "Passive" immunity means the person is receiving pre-made antibodies, rather than their own immune system working to make them (which would be active immunity, like surviving a bacterial infection or taking a vaccine).
Step 2: "Natural" means the immunity was acquired through natural life processes, not via clinical intervention (like an injection of antivenom).
Step 3: The transfer of IgG antibodies across the placenta to a fetus, or IgA antibodies via breast milk to a nursing infant, is the prime example of natural passive immunity.

Detailed Solution:
Step 1: Predicting the 3D structure of a protein from its amino acid sequence is very difficult ab initio (from scratch).
Step 2: The most reliable and common method is homology modeling (comparative modeling).
Step 3: This approach is based on the evolutionary principle that protein structure is highly conserved. If two proteins share significant sequence identity (they are close homologs), they will fold into virtually the same 3D structure.

Detailed Solution:
Step 1: According to the Central Dogma of Molecular Biology in eukaryotes, the flow of genetic information starts in the nucleus with transcription (2) of DNA into pre-mRNA.
Step 2: The pre-mRNA undergoes post-transcriptional modifications, primarily splicing (4) to remove introns, forming mature mRNA.
Step 3: The mRNA enters the cytoplasm where it undergoes translation (3) by a ribosome.
Step 4: Translation yields a raw polypeptide chain, which is the unformed protein (1).
Step 5: Finally, this chain undergoes protein folding (5) into its functional 3D conformation.
Result: The correct sequence is 2 → 4 → 3 → 1 → 5.

Detailed Solution:
Step 1: Simple proteins are composed exclusively of amino acids.
Step 2: A conjugated protein is a complex protein that requires a non-amino acid component (a prosthetic group) attached to it to function properly.
Step 3: A metalloprotein (e.g., hemoglobin, which contains Iron) is a prime example of a conjugated protein because it contains a metal ion prosthetic group strongly bound to the polypeptide chain.

Detailed Solution:
Step 1: During mRNA translation by the ribosome, the reading of codons begins at a specific initiation signal.
Step 2: In almost all organisms, this universal start codon is AUG.
Step 3: The AUG codon codes for the amino acid Methionine (or formyl-methionine in prokaryotes), ensuring that nearly all newly synthesized proteins begin with Methionine.

Detailed Solution:
Step 1: When a neuron is not transmitting a signal, it is "at rest."
Step 2: Due to the action of the sodium-potassium pump (which pumps 3 $Na^+$ out for every 2 $K^+$ in) and the presence of negatively charged proteins inside the cell, the interior of the neuron is more negative relative to the outside.
Step 3: This creates an electrical voltage difference across the membrane known as the resting membrane potential, which is typically measured at about -70 millivolts (-70 mV) in humans.

Detailed Solution:
Step 1: An Artificial Neural Network (ANN) is undeniably a machine learning tool. However, the phrasing "performs like a..." refers to its biological inspiration.
Step 2: ANNs are computational architectures explicitly designed to mimic the way the biological human brain (and human cognition) processes information—using interconnected nodes (artificial neurons) that learn from experience.
Step 3: In the context of multiple-choice questions assessing the biological inspiration of ANNs, it is said to perform like a human (brain) in recognizing complex patterns.

Detailed Solution:
Step 1: In an Artificial Neural Network, inputs are multiplied by a numerical weight, which dictates the strength of the connection between two nodes.
Step 2: In a biological nervous system, the physical connection between two neurons is the synapse.
Step 3: A biological synapse can be strengthened or weakened (synaptic plasticity), making it easier or harder to pass a signal. Therefore, the artificial "weight" is the direct equivalent of biological synaptic strength.

Detailed Solution:
Step 1: A typical multipolar neuron (nerve cell) has three main parts: dendrites, soma (cell body), and an axon.
Step 2: The cell receives incoming signals through highly branched, tree-like structures called dendrites. A single neuron usually has many dendrites to receive signals from thousands of neighbors.
Step 3: The processed signal is then sent away to the next cell via the axon. A typical neuron possesses only one axon.

Detailed Solution:
Step 1: Inactivated vaccines contain completely killed pathogens. Subunit and recombinant vaccines contain only isolated pieces (proteins) of the pathogen.
Step 2: A live-attenuated vaccine contains the actual living pathogen, but it has been cultured in a lab over generations to drastically reduce its disease-causing ability (virulence).
Step 3: Because it is alive, it closely mimics a natural infection and provokes a very strong and lasting immune response without causing serious illness (e.g., MMR or OPV vaccines).

Detailed Solution:
Step 1: Bepitope is obviously for B-cell epitopes. BLAST is a general sequence alignment tool.
Step 2: To predict T-cell epitopes, an algorithm must predict whether a peptide sequence will successfully bind to an MHC (Major Histocompatibility Complex) molecule to be presented to a T-cell.
Step 3: NetMHCpan is the current gold-standard, state-of-the-art bioinformatics server that uses artificial neural networks to predict the binding of peptides to any known MHC molecule, thereby predicting T-cell epitopes. (Note: TEPITOPE is also a valid historical T-cell epitope predictor, but NetMHCpan is the most universally taught modern tool for this purpose).

Detailed Solution:
Step 1: Chicken pox (Varicella), whooping cough (Pertussis), and diphtheria all have highly effective, standard childhood vaccines available globally (e.g., DTaP vaccine).
Step 2: Malaria is caused by the Plasmodium parasite. Because it is a complex eukaryotic parasite with a multi-stage life cycle and high mutation rate, creating a vaccine has been notoriously difficult.
Step 3: Historically, and in the context of standard biology curricula, malaria is taught as the classic example of a devastating global disease that lacks a universally deployed, highly effective vaccine (though the RTS,S vaccine has recently seen limited rollout in Africa, it is not widely available like the others).

1. Introduction to Genetic Algorithms (GA):
A Genetic Algorithm is a computational metaheuristic search and optimization technique inspired by Charles Darwin's theory of natural evolution. Developed largely by John Holland in the 1970s, it belongs to the larger class of Evolutionary Algorithms (EA). GAs are used to find optimal or near-optimal solutions to complex, non-linear problems where traditional algorithms fail or take too much time.

2. Basic Terminology:

• Population: A set of potential solutions.
• Chromosome: A single potential solution within the population (often represented as a string of binary 1s and 0s).
• Gene: A single bit/variable within the chromosome.
• Fitness Function: A mathematical function that evaluates how "good" a chromosome (solution) is at solving the problem.

3. The Five Phases of the Algorithm:
Step 1: Initialization: The algorithm randomly generates an initial population of chromosomes (potential solutions).
Step 2: Fitness Evaluation: The fitness function scores every chromosome in the population. Higher scores mean better solutions.
Step 3: Selection: Based on the "survival of the fittest" principle, the algorithm selects the best-performing chromosomes to be "parents" for the next generation. (Methods include Roulette Wheel selection or Tournament selection).
Step 4: Crossover (Recombination): The selected parents are paired up. A random crossover point is chosen, and they swap genetic information (bits) to produce "offspring." This ensures the offspring inherit good traits from both parents.
Step 5: Mutation: To maintain genetic diversity and prevent the algorithm from getting stuck in a local optimum, a few random bits in the offspring's chromosomes are flipped (mutated) with a very low probability (e.g., 1%).

4. Termination:
The new offspring replace the old population, and Steps 2-5 repeat. The loop terminates when a satisfactory solution is found, or a maximum number of generations is reached.

5. Applications in Bioinformatics:
GAs are heavily used in biology for DNA fragment assembly, Multiple Sequence Alignment (MSA), protein folding prediction, and drug design (docking simulations).

1. Introduction to Carbohydrates:
Carbohydrates are the most abundant biomolecules on Earth. They are organic compounds made of Carbon, Hydrogen, and Oxygen, generally following the empirical formula $(CH_2O)_n$. They serve two primary functions: providing immediate/stored energy, and acting as structural frameworks for cells.

2. Classification of Carbohydrates:
Carbohydrates are classified based on the number of sugar (saccharide) units they contain.

A. Monosaccharides (Simple Sugars):
These are the simplest carbohydrates and cannot be hydrolyzed into smaller units. They are the building blocks.

• Example: Glucose ($C_6H_12O_6$). The primary energy source for cellular respiration in animals.
• Example: Fructose (Fruit sugar). Has the same formula as glucose but a different structural arrangement (isomer).
• Example: Ribose ($C_5H_10O_5$). The structural backbone of RNA.

B. Disaccharides:
Formed when two monosaccharides undergo a dehydration synthesis reaction, joining via a covalent glycosidic bond.

• Example: Sucrose (Table sugar). Formed by linking Glucose + Fructose. It is the main transport sugar in plants.
• Example: Lactose (Milk sugar). Formed by linking Glucose + Galactose.
• Example: Maltose (Malt sugar). Formed by linking Glucose + Glucose.

C. Oligosaccharides:
Short chains of 3 to 10 monosaccharides. They are often found attached to proteins (glycoproteins) or lipids (glycolipids) on the cell membrane, serving crucially in cell-cell recognition and signaling.

D. Polysaccharides (Complex Carbohydrates):
Long polymer chains containing hundreds or thousands of monosaccharides. They serve for storage or structure.

• Storage Examples: - Starch: How plants store excess glucose (e.g., in potatoes). - Glycogen: How animals store excess glucose (stored heavily in the liver and muscles).
• Structural Examples: - Cellulose: Forms the tough cell wall of plants. Humans cannot digest it (dietary fiber). - Chitin: Forms the exoskeleton of insects/crustaceans and the cell walls of fungi.

1. What is Secondary Structure Prediction?
Protein secondary structure (SS) prediction takes a 1D amino acid sequence and computationally predicts which local regions will fold into specific SS elements: Alpha-helices (H), Beta-sheets (E), or Random Coils (C). This is a critical intermediate step in bioinformatics toward predicting the full 3D tertiary structure and understanding protein function.

2. Evolution and Types of Prediction Tools:
First Generation (Statistical Methods):
These algorithms relied entirely on single-residue statistics derived from early X-ray crystallography databases. They calculated the propensity (likelihood) of a specific amino acid to appear in a helix or sheet.

• Chou-Fasman Method: Established rules based on individual amino acid preferences (e.g., Proline breaks helices). Accuracy was only about 50-60%.
• GOR Method (Garnier-Osguthorpe-Robson): Improved upon Chou-Fasman by considering a "sliding window" of 17 amino acids, recognizing that neighboring residues influence folding. Achieved ~65% accuracy.

Second Generation (Machine Learning Methods):
With the advent of Artificial Neural Networks (ANN), algorithms were trained on larger datasets of known structures.

• PHD (Profile network from HeiDelberg): Used sliding windows fed into neural networks. It recognized patterns better than simple statistics, pushing accuracy to ~70%.

Third Generation (Evolutionary Profile Methods - The Modern Standard):
The major breakthrough was realizing that evolutionary conservation dictates structure. Modern tools first run a PSI-BLAST search against sequence databases to find all related proteins. They create a Position-Specific Scoring Matrix (PSSM), which shows the evolutionary profile of the sequence. This rich evolutionary data is then fed into neural networks.

• PSIPRED: Uses a strict two-stage feed-forward neural network analyzing PSI-BLAST profiles. It achieves high accuracy (over 80%) and is widely used today.
• Jpred / Jnet: Uses Hidden Markov Models (HMM) and neural networks combined with evolutionary profiles. It offers a user-friendly web server interface for high-accuracy predictions.

3. Evaluation of Tools:
The accuracy of these tools is measured using the Q3 score, which represents the percentage of residues correctly predicted as either Helix, Sheet, or Coil. Modern third-generation tools consistently score between 80-85%.

1. Introduction to Translation:
Translation is the highly coordinated biological process by which a cell synthesizes proteins. The ribosome reads the genetic information carried by messenger RNA (mRNA) in three-letter nucleotide blocks called codons, and uses transfer RNA (tRNA) to string together the corresponding amino acids into a polypeptide chain.

2. Components Required:
- mRNA: The instruction manual.
- Ribosome: The workbench (composed of a small 30S/40S subunit and a large 50S/60S subunit).
- tRNA: The delivery trucks carrying specific amino acids, equipped with an anticodon that matches the mRNA codon.
- Amino Acids & ATP/GTP: Building blocks and energy.

3. The Three Stages of Translation:
Stage 1: Initiation
The small ribosomal subunit binds to the mRNA and scans for the start codon (AUG). An initiator tRNA carrying the amino acid Methionine (Met) binds its anticodon (UAC) to the start codon. Finally, the large ribosomal subunit attaches, completing the initiation complex. The initiator tRNA sits in the ribosome's 'P-site'.

Stage 2: Elongation
The ribosome has three slots: A (Acceptor), P (Peptidyl), and E (Exit).

• A new tRNA matching the next mRNA codon enters the A-site.
• The ribosome catalyzes a peptide bond, transferring the growing amino acid chain from the tRNA in the P-site to the new amino acid in the A-site.
• Translocation: The ribosome shifts forward exactly one codon (3 nucleotides). The empty tRNA moves to the E-site and leaves. The tRNA holding the growing chain moves from the A-site to the P-site, opening the A-site for the next tRNA. This cycle repeats rapidly.

Stage 3: Termination
Elongation continues until a Stop Codon (UAA, UAG, or UGA) enters the A-site. There are no tRNAs for stop codons. Instead, a "Release Factor" protein enters the A-site. This triggers the ribosome to sever the bond holding the polypeptide to the final tRNA. The newly synthesized protein is released, and the ribosomal subunits disassemble.

4. Diagrammatic Representation:
(Note to student: Draw a ribosome showing the A, P, and E sites, the mRNA threading through it, and the incoming/outgoing tRNAs as depicted below).

       [Growing Polypeptide Chain]
            |      |      |
          (Met)--(Ala)--(Lys) 
                   |
                 [tRNA] <---- (Peptide bond forming)
                   |
    E-site       P-site       A-site     [Incoming tRNA carrying Proline]
 +---------+  +---------+  +---------+
 |  Empty  |  |   UUC   |  |         |   <- Ribosome Large Subunit
 |  tRNA   |  | (Anti)  |  | (Anti)  |
 +---------+  +---------+  +---------+
=========================================
 ...  AUG        AAG          CCG  ...   <- mRNA strand
=========================================
              

1. What is an Artificial Neural Network (ANN)?
An Artificial Neural Network is a powerful machine learning framework inspired by the biological architecture of the human brain. Just as a brain consists of billions of interconnected biological neurons, an ANN consists of layers of interconnected artificial nodes (called perceptrons or artificial neurons). ANNs excel at finding hidden patterns in complex, non-linear data without being explicitly programmed with rules.

2. Architecture of an ANN:
A standard feed-forward ANN consists of three main types of layers:

• Input Layer: Receives the raw data (e.g., pixels of an image, or patient medical stats). Each node represents one feature.
• Hidden Layer(s): One or more layers between input and output where the actual computational "thinking" occurs. The nodes here extract abstract features.
• Output Layer: Produces the final prediction or classification (e.g., "Malignant Tumor" vs "Benign Tumor").

3. How does it work? (The Math made simple):
Every connection between two nodes has a Weight ($w$), representing the connection's strength (synapse). Each node also has a Bias ($b$). When a node receives data, it performs a simple calculation:

$$ Output = ActivationFunction( \sum (Input imes Weight) + Bias ) $$
The Activation Function (like ReLU or Sigmoid) decides if the node should "fire" its signal to the next layer, mimicking an action potential. During training, an algorithm called Backpropagation checks the network's errors and continuously adjusts the weights until the predictions become accurate.

4. Real-World Example in Biology/Medicine:
Scenario: Diagnosing Skin Cancer.

• Input: You feed a 100x100 pixel image of a skin mole into the ANN. The network has 10,000 input nodes (one for the color/intensity of each pixel).
• Hidden Layers: The first hidden layer might learn to detect simple edges. The second layer combines edges to detect shapes. The third layer detects complex patterns like irregular borders or specific color blotches indicative of melanoma.
• Output: The final layer has two nodes: "Melanoma" and "Normal." Based on the calculations through the hidden layers, one node fires stronger.
• Result: The ANN successfully classifies the image, acting as an automated, highly accurate diagnostic tool.

1. Introduction to the Immune System:
The human immune system is a highly complex network of cells and proteins defending the body against infection. It is broadly divided into two interdependent branches: the fast-acting Innate Immunity and the highly specific Acquired (Adaptive) Immunity.

2. Innate Immunity (The First Responder):
Innate immunity is present from birth. It provides a generalized, immediate defense against a wide array of pathogens. It does not recognize specific strains of viruses or bacteria, nor does it retain a memory of past infections.

• Physical/Chemical Barriers: The skin (blocks entry), stomach acid (destroys swallowed bacteria), and mucus/tears (trap and wash away pathogens).
• Cellular Defenses: Phagocytic white blood cells (like Macrophages and Neutrophils) patrol the body, eating and destroying any generic foreign material they encounter. Natural Killer (NK) cells indiscriminately destroy virally infected host cells.
• Inflammatory Response: Swelling, redness, and fever to create a hostile environment for pathogens and recruit immune cells.

3. Acquired / Adaptive Immunity (The Elite Specialists):
Acquired immunity develops over a lifetime as the body is exposed to diseases (or vaccines). It is highly specific; it tailors a unique biochemical weapon for every distinct pathogen it encounters. Crucially, it possesses immunological memory.

• Cell-Mediated Immunity: Driven by T-lymphocytes (T-cells). Cytotoxic T-cells precisely target and destroy host cells infected by a specific virus.
• Humoral Immunity: Driven by B-lymphocytes (B-cells). When activated, they become plasma cells that produce massive amounts of highly specific Antibodies that lock onto free-floating pathogens to neutralize them.
• Memory: Memory B and T cells remain in the body for decades, ensuring that a second exposure to the exact same pathogen is neutralized before symptoms appear.

4. Key Differences (Comparison):

Feature	Innate Immunity	Acquired (Adaptive) Immunity
Response Time	Immediate (Minutes to hours)	Slow (Takes days to weeks to fully activate)
Specificity	Non-specific (Attacks any foreign body)	Highly specific (Targets exact molecular antigens)
Memory	No memory generated	Creates long-lasting immunological memory
Key Cells	Macrophages, Neutrophils, NK Cells	T-Cells and B-Cells

1. Introduction:
The development of vaccines for the SARS-CoV-2 virus (the cause of COVID-19) was a historic biomedical achievement. Normally taking 10-15 years, these vaccines were developed, tested, and authorized in under a year. This was possible due to decades of prior research into mRNA technology, massive global funding, and the early genomic sequencing of the virus by Chinese scientists in January 2020.

2. Target Identification: The Spike Protein:
Scientists identified the "Spike (S) Protein" on the surface of SARS-CoV-2 as the prime target. The virus uses this spike to bind to ACE2 receptors on human lung cells to force entry. By training the immune system to recognize and attack the Spike protein, neutralizing antibodies could physically block the virus from entering host cells, preventing infection.

3. The Four Main Vaccine Technologies Developed:
A. mRNA Vaccines (Pfizer-BioNTech & Moderna):
- Mechanism: Scientists synthesized a strip of messenger RNA containing the genetic instructions to build the Spike protein. This mRNA is fragile, so it was encased in a protective Lipid Nanoparticle (LNP).
- Action: Once injected, the LNPs fuse with host cells. The host's own ribosomes read the mRNA and manufacture harmless Spike proteins, displaying them to the immune system. The mRNA is then degraded.
- Advantage: Extremely fast to design and manufacture; highly effective (90-95% initial efficacy).

B. Viral Vector Vaccines (Oxford-AstraZeneca & J&J):
- Mechanism: Scientists took a harmless, modified Adenovirus (a common cold virus that cannot replicate in humans) and inserted the DNA code for the SARS-CoV-2 Spike protein into it.
- Action: The adenovirus acts as a Trojan horse, delivering the Spike DNA into the nucleus of human cells. The cell transcribes it to mRNA, translates it to Spike protein, and triggers immunity.
- Advantage: Can be stored at normal refrigerator temperatures (unlike the ultra-cold storage required for mRNA initially).

C. Protein Subunit Vaccines (Novavax):
- Mechanism: Instead of making the human body produce the Spike protein, scientists used moth cells in a lab bioreactor to mass-produce the Spike proteins directly.
- Action: These purified proteins are injected into the patient along with an "adjuvant" (a chemical that irritates the immune system to ensure a strong reaction).
- Advantage: This is a very traditional, proven technology (used for Hepatitis B vaccines) with a strong safety profile.

D. Inactivated Virus Vaccines (Sinovac & Covaxin):
- Mechanism: The actual SARS-CoV-2 virus is grown in huge quantities in secure bio-labs, and then completely killed (inactivated) using chemicals like beta-propiolactone.
- Action: The dead virus is injected. It cannot cause disease, but the immune system recognizes the full array of viral proteins (not just the spike) and mounts a defense.
- Advantage: Extremely well-established technology (used for polio and flu vaccines).

4. Clinical Trials and Efficacy:
Regardless of the technology, every vaccine had to pass strict Phase I (safety), Phase II (dosing/immune response), and massive Phase III (tens of thousands of volunteers to prove efficacy against placebo) trials before receiving Emergency Use Authorization (EUA) from global health agencies.

1. Introduction and Algorithm Choice:
Given a database of raw medical images (e.g., histology slides of cells) to predict the presence of cancer, traditional machine learning algorithms (like Decision Trees or SVMs) perform poorly because they require humans to manually extract features (like cell size, shape, and color).
For raw image data, the absolute gold-standard, state-of-the-art machine learning algorithm is a deep learning architecture known as a Convolutional Neural Network (CNN).

2. What is a Convolutional Neural Network (CNN)?
A CNN is a highly specialized Artificial Neural Network designed specifically to process data that has a grid-like topology, such as an image (a grid of pixels). Unlike standard networks, CNNs can automatically learn spatial hierarchies of features directly from raw images.

3. Step-by-Step Architecture and Process for Cancer Prediction:
Step A: Data Preprocessing & Input
The database of cancer cell images must first be labeled by expert pathologists (e.g., "Malignant" vs "Benign"). The images are resized to a uniform dimension (e.g., $256 imes 256$ pixels), and the pixel color values are normalized (scaled between 0 and 1).

Step B: The Convolutional Layers (Feature Extraction)
Instead of looking at every pixel individually, the network applies "filters" or "kernels" (small grids, e.g., $3 imes 3$ pixels) that slide (convolve) across the entire image.

• Early Layers: The first convolutional layers learn to detect simple, low-level features like edges, lines, and basic color gradients in the cell images.
• Deep Layers: As the data passes deeper into the network, the layers combine these simple edges to detect complex, high-level biological features indicative of cancer—such as enlarged nuclei, irregular cellular borders, and mitotic figures (cells dividing abnormally).

Step C: Pooling Layers (Downsampling)
After a convolution, a pooling layer (usually Max Pooling) is applied. This reduces the spatial dimensions of the image, keeping only the most important features. This drastically reduces the computational load and makes the algorithm robust to slight shifts or rotations of the cancer cell in the image.

Step D: Fully Connected Layers (Classification)
After passing through multiple alternating Convolution and Pooling layers, the 2D grid of extracted features is flattened into a 1D vector of numbers. This vector is fed into a standard Artificial Neural Network (Fully Connected Layers).
These final layers act as the "brain," weighing all the complex cellular features extracted by the convolutions to make a final decision.

Step E: Output Layer (Prediction)
The final layer usually uses a Softmax or Sigmoid activation function. It outputs a probability score between 0 and 1. For example, it might output a 0.92 probability that the given cell image is malignant, successfully predicting cancer.

4. Training the Model (Backpropagation):
Initially, the CNN will guess randomly. During training, it compares its prediction to the pathologist's true label. It calculates the error (loss function) and uses an optimization algorithm (like Adam or Stochastic Gradient Descent) combined with Backpropagation to adjust the weights of all its filters. After analyzing thousands of images repeatedly (epochs), the CNN becomes highly accurate, often matching or exceeding human expert performance.