{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.11.13","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"none","dataSources":[{"sourceId":116062,"databundleVersionId":14875579,"sourceType":"competition"},{"sourceId":14173811,"sourceType":"datasetVersion","datasetId":9034760}],"dockerImageVersionId":31192,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"
\n

Ozan M.

\n

Data Scientist / Data Analyst

\n
\n
\n\n\n\n\n\n\n\n\n\n
\n
\n
","metadata":{}},{"cell_type":"markdown","source":"
\n

CAFA-6 Protein Function Prediction

\n

Advanced Statistical Analysis & Transformer-Based Deep Learning

\n
","metadata":{}},{"cell_type":"markdown","source":"
\n

🚀 Environment Configuration

\n\n
","metadata":{}},{"cell_type":"code","source":"import numpy as np\nimport pandas as pd\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nfrom pathlib import Path\nimport warnings\nfrom collections import Counter, defaultdict\nimport re\n\nwarnings.filterwarnings('ignore')\n\nCOLORS = {\n 'bg': '#0a1929',\n 'surface': '#1a2332', \n 'primary': '#60a5fa',\n 'secondary': '#38bdf8',\n 'accent': '#818cf8',\n 'text': '#f1f5f9',\n 'muted': '#94a3b8'\n}\n\nplt.rcParams.update({\n 'figure.facecolor': COLORS['bg'],\n 'axes.facecolor': COLORS['surface'],\n 'axes.edgecolor': COLORS['primary'],\n 'axes.labelcolor': COLORS['text'],\n 'text.color': COLORS['text'],\n 'xtick.color': COLORS['text'],\n 'ytick.color': COLORS['text'],\n 'grid.color': COLORS['muted'],\n 'grid.alpha': 0.2,\n 'figure.figsize': (15, 8),\n 'font.size': 11,\n 'axes.labelsize': 12,\n 'axes.titlesize': 14\n})\n\ndef add_watermark(ax):\n ax.text(0.98, 0.02, 'Created by Ozan M.', \n transform=ax.transAxes, \n fontsize=9, \n color=COLORS['muted'], \n alpha=0.7,\n ha='right', \n va='bottom',\n style='italic')\n\nBASE_DIR = Path('/kaggle/input/cafa-6-protein-function-prediction')\nTRAIN_DIR = BASE_DIR / 'Train'","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T08:55:32.442149Z","iopub.execute_input":"2025-12-17T08:55:32.442419Z","iopub.status.idle":"2025-12-17T08:55:36.688916Z","shell.execute_reply.started":"2025-12-17T08:55:32.442389Z","shell.execute_reply":"2025-12-17T08:55:36.688046Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

📂 Data Loading & Integration

\n\n
","metadata":{}},{"cell_type":"code","source":"FILES = {\n 'train_seq': TRAIN_DIR / 'train_sequences.fasta',\n 'train_terms': TRAIN_DIR / 'train_terms.tsv',\n 'train_tax': TRAIN_DIR / 'train_taxonomy.tsv',\n 'ia': BASE_DIR / 'IA.tsv',\n 'obo': TRAIN_DIR / 'go-basic.obo'\n}\n\ndef load_fasta(filepath):\n sequences = {}\n current_id = None\n current_seq = []\n \n with open(filepath, 'r') as f:\n for line in f:\n line = line.strip()\n if line.startswith('>'):\n if current_id:\n sequences[current_id] = ''.join(current_seq)\n current_id = line[1:].split()[0]\n current_seq = []\n else:\n current_seq.append(line)\n if current_id:\n sequences[current_id] = ''.join(current_seq)\n \n return sequences\n\nsequences = load_fasta(FILES['train_seq'])\ntrain_terms = pd.read_csv(FILES['train_terms'], sep='\\t', header=None, names=['EntryID', 'GO_Term', 'Aspect'])\ntrain_tax = pd.read_csv(FILES['train_tax'], sep='\\t', header=None, names=['EntryID', 'Taxonomy'])\nia_scores = pd.read_csv(FILES['ia'], sep='\\t', header=None, names=['GO_Term', 'IA_Score'])\n\nseq_df = pd.DataFrame([\n {'EntryID': k, 'Sequence': v, 'Length': len(v)} \n for k, v in sequences.items()\n])\n\ndata = seq_df.merge(train_terms, on='EntryID', how='left')\ndata = data.merge(train_tax, on='EntryID', how='left')\ndata = data.merge(ia_scores, on='GO_Term', how='left')\n\ndata.head()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T08:55:36.690933Z","iopub.execute_input":"2025-12-17T08:55:36.691434Z","iopub.status.idle":"2025-12-17T08:55:38.381818Z","shell.execute_reply.started":"2025-12-17T08:55:36.691403Z","shell.execute_reply":"2025-12-17T08:55:38.380910Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

📊 GO Ontology Distribution Analysis

\n\n
","metadata":{}},{"cell_type":"code","source":"aspect_counts = train_terms['Aspect'].value_counts()\ngo_term_counts = train_terms.groupby('Aspect')['GO_Term'].nunique()\nproteins_per_aspect = train_terms.groupby('Aspect')['EntryID'].nunique()\n\nfig, axes = plt.subplots(1, 3, figsize=(18, 6))\nfig.patch.set_facecolor(COLORS['bg'])\n\ncolors_map = {'MF': '#60a5fa', 'BP': '#38bdf8', 'CC': '#818cf8'}\naspect_colors = [colors_map.get(x, COLORS['primary']) for x in aspect_counts.index]\n\nbars1 = axes[0].bar(aspect_counts.index, aspect_counts.values, \n color=aspect_colors, edgecolor='none', alpha=0.85, width=0.6)\nfor bar in bars1:\n bar.set_linewidth(0)\naxes[0].set_xlabel('GO Aspect', fontsize=12, color=COLORS['text'])\naxes[0].set_ylabel('Annotations Count', fontsize=12, color=COLORS['text'])\naxes[0].set_title('Annotation Distribution', fontsize=14, fontweight='600', \n color=COLORS['primary'], pad=20)\naxes[0].spines['top'].set_visible(False)\naxes[0].spines['right'].set_visible(False)\naxes[0].spines['left'].set_color(COLORS['muted'])\naxes[0].spines['bottom'].set_color(COLORS['muted'])\naxes[0].tick_params(colors=COLORS['text'])\naxes[0].grid(axis='y', alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[0])\n\nbars2 = axes[1].bar(go_term_counts.index, go_term_counts.values,\n color=aspect_colors, edgecolor='none', alpha=0.85, width=0.6)\nfor bar in bars2:\n bar.set_linewidth(0)\naxes[1].set_xlabel('GO Aspect', fontsize=12, color=COLORS['text'])\naxes[1].set_ylabel('Unique Terms', fontsize=12, color=COLORS['text'])\naxes[1].set_title('GO Term Vocabulary', fontsize=14, fontweight='600', \n color=COLORS['primary'], pad=20)\naxes[1].spines['top'].set_visible(False)\naxes[1].spines['right'].set_visible(False)\naxes[1].spines['left'].set_color(COLORS['muted'])\naxes[1].spines['bottom'].set_color(COLORS['muted'])\naxes[1].tick_params(colors=COLORS['text'])\naxes[1].grid(axis='y', alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[1])\n\nbars3 = axes[2].bar(proteins_per_aspect.index, proteins_per_aspect.values,\n color=aspect_colors, edgecolor='none', alpha=0.85, width=0.6)\nfor bar in bars3:\n bar.set_linewidth(0)\naxes[2].set_xlabel('GO Aspect', fontsize=12, color=COLORS['text'])\naxes[2].set_ylabel('Unique Proteins', fontsize=12, color=COLORS['text'])\naxes[2].set_title('Protein Coverage', fontsize=14, fontweight='600', \n color=COLORS['primary'], pad=20)\naxes[2].spines['top'].set_visible(False)\naxes[2].spines['right'].set_visible(False)\naxes[2].spines['left'].set_color(COLORS['muted'])\naxes[2].spines['bottom'].set_color(COLORS['muted'])\naxes[2].tick_params(colors=COLORS['text'])\naxes[2].grid(axis='y', alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[2])\n\nplt.tight_layout()\nplt.show()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T08:55:38.382812Z","iopub.execute_input":"2025-12-17T08:55:38.383095Z","iopub.status.idle":"2025-12-17T08:55:39.226006Z","shell.execute_reply.started":"2025-12-17T08:55:38.383072Z","shell.execute_reply":"2025-12-17T08:55:39.225099Z"},"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

🧬 Sequence Characteristics Analysis

\n\n
","metadata":{}},{"cell_type":"code","source":"seq_lengths = [len(v) for v in sequences.values()]\nseq_stats = pd.Series(seq_lengths).describe()\n\nfig, axes = plt.subplots(2, 2, figsize=(18, 12))\nfig.patch.set_facecolor(COLORS['bg'])\n\naxes[0, 0].hist(seq_lengths, bins=80, color=COLORS['primary'], \n edgecolor='none', alpha=0.85)\naxes[0, 0].axvline(np.median(seq_lengths), color=COLORS['secondary'], \n linestyle='--', linewidth=2, label=f'Median: {np.median(seq_lengths):.0f}')\naxes[0, 0].set_xlabel('Sequence Length', fontsize=12, color=COLORS['text'])\naxes[0, 0].set_ylabel('Frequency', fontsize=12, color=COLORS['text'])\naxes[0, 0].set_title('Protein Sequence Length Distribution', fontsize=14, \n fontweight='600', color=COLORS['primary'], pad=20)\naxes[0, 0].legend(fontsize=10, loc='upper right', framealpha=0.9)\naxes[0, 0].spines['top'].set_visible(False)\naxes[0, 0].spines['right'].set_visible(False)\naxes[0, 0].spines['left'].set_color(COLORS['muted'])\naxes[0, 0].spines['bottom'].set_color(COLORS['muted'])\naxes[0, 0].grid(axis='y', alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[0, 0])\n\nbp = axes[0, 1].boxplot([seq_lengths], vert=True, patch_artist=True, widths=0.5,\n boxprops=dict(facecolor=COLORS['primary'], alpha=0.85, edgecolor='none'),\n medianprops=dict(color=COLORS['secondary'], linewidth=2.5),\n whiskerprops=dict(color=COLORS['muted'], linewidth=1.5),\n capprops=dict(color=COLORS['muted'], linewidth=1.5),\n flierprops=dict(marker='o', markerfacecolor=COLORS['accent'], \n markersize=4, alpha=0.5, markeredgecolor='none'))\naxes[0, 1].set_ylabel('Sequence Length', fontsize=12, color=COLORS['text'])\naxes[0, 1].set_title('Length Distribution Summary', fontsize=14, \n fontweight='600', color=COLORS['primary'], pad=20)\naxes[0, 1].set_xticklabels(['All Proteins'], fontsize=11)\naxes[0, 1].spines['top'].set_visible(False)\naxes[0, 1].spines['right'].set_visible(False)\naxes[0, 1].spines['left'].set_color(COLORS['muted'])\naxes[0, 1].spines['bottom'].set_color(COLORS['muted'])\naxes[0, 1].grid(axis='y', alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[0, 1])\n\nall_aa = ''.join(sequences.values())\naa_counts = Counter(all_aa)\naa_freq = pd.Series(aa_counts).sort_values(ascending=False)\n\naxes[1, 0].bar(range(len(aa_freq)), aa_freq.values, \n color=COLORS['primary'], edgecolor='none', alpha=0.85, width=0.8)\naxes[1, 0].set_xticks(range(len(aa_freq)))\naxes[1, 0].set_xticklabels(aa_freq.index, fontsize=10)\naxes[1, 0].set_xlabel('Amino Acid', fontsize=12, color=COLORS['text'])\naxes[1, 0].set_ylabel('Count', fontsize=12, color=COLORS['text'])\naxes[1, 0].set_title('Amino Acid Frequency Distribution', fontsize=14, \n fontweight='600', color=COLORS['primary'], pad=20)\naxes[1, 0].spines['top'].set_visible(False)\naxes[1, 0].spines['right'].set_visible(False)\naxes[1, 0].spines['left'].set_color(COLORS['muted'])\naxes[1, 0].spines['bottom'].set_color(COLORS['muted'])\naxes[1, 0].grid(axis='y', alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[1, 0])\n\nannotations_per_protein = train_terms.groupby('EntryID').size()\naxes[1, 1].hist(annotations_per_protein.values, bins=60, color=COLORS['secondary'], \n edgecolor='none', alpha=0.85)\naxes[1, 1].axvline(annotations_per_protein.median(), color=COLORS['accent'], \n linestyle='--', linewidth=2, label=f'Median: {annotations_per_protein.median():.0f}')\naxes[1, 1].set_xlabel('Annotations per Protein', fontsize=12, color=COLORS['text'])\naxes[1, 1].set_ylabel('Frequency', fontsize=12, color=COLORS['text'])\naxes[1, 1].set_title('GO Term Annotations per Protein', fontsize=14, \n fontweight='600', color=COLORS['primary'], pad=20)\naxes[1, 1].legend(fontsize=10, loc='upper right', framealpha=0.9)\naxes[1, 1].spines['top'].set_visible(False)\naxes[1, 1].spines['right'].set_visible(False)\naxes[1, 1].spines['left'].set_color(COLORS['muted'])\naxes[1, 1].spines['bottom'].set_color(COLORS['muted'])\naxes[1, 1].grid(axis='y', alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[1, 1])\n\nplt.tight_layout()\nplt.show()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T08:55:39.226896Z","iopub.execute_input":"2025-12-17T08:55:39.227179Z","iopub.status.idle":"2025-12-17T08:55:43.292992Z","shell.execute_reply.started":"2025-12-17T08:55:39.227149Z","shell.execute_reply":"2025-12-17T08:55:43.292165Z"},"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

🎯 GO Term Frequency Analysis

\n\n
","metadata":{}},{"cell_type":"code","source":"top_terms = train_terms['GO_Term'].value_counts().head(20)\nterm_aspect_map = train_terms.groupby('GO_Term')['Aspect'].first()\ntop_term_aspects = term_aspect_map[top_terms.index]\n\ncolors_for_terms = [colors_map.get(aspect, COLORS['primary']) for aspect in top_term_aspects]\n\nfig, ax = plt.subplots(figsize=(16, 8))\nfig.patch.set_facecolor(COLORS['bg'])\n\nbars = ax.barh(range(len(top_terms)), top_terms.values, \n color=colors_for_terms, edgecolor='none', alpha=0.85, height=0.7)\n\nax.set_yticks(range(len(top_terms)))\nax.set_yticklabels(top_terms.index, fontsize=10)\nax.set_xlabel('Frequency', fontsize=12, color=COLORS['text'])\nax.set_ylabel('GO Term', fontsize=12, color=COLORS['text'])\nax.set_title('Top 20 Most Frequent GO Terms', fontsize=14, \n fontweight='600', color=COLORS['primary'], pad=20)\nax.spines['top'].set_visible(False)\nax.spines['right'].set_visible(False)\nax.spines['left'].set_color(COLORS['muted'])\nax.spines['bottom'].set_color(COLORS['muted'])\nax.grid(axis='x', alpha=0.15, linestyle='-', linewidth=0.5)\nax.invert_yaxis()\n\nlegend_elements = [plt.Rectangle((0,0),1,1, facecolor=colors_map['MF'], alpha=0.85, label='MF'),\n plt.Rectangle((0,0),1,1, facecolor=colors_map['BP'], alpha=0.85, label='BP'),\n plt.Rectangle((0,0),1,1, facecolor=colors_map['CC'], alpha=0.85, label='CC')]\nax.legend(handles=legend_elements, loc='lower right', fontsize=10, framealpha=0.9)\n\nadd_watermark(ax)\nplt.tight_layout()\nplt.show()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T08:55:43.294077Z","iopub.execute_input":"2025-12-17T08:55:43.294867Z","iopub.status.idle":"2025-12-17T08:55:43.936196Z","shell.execute_reply.started":"2025-12-17T08:55:43.294831Z","shell.execute_reply":"2025-12-17T08:55:43.935390Z"},"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

⚖️ Information Accretion Analysis

\n\n
","metadata":{}},{"cell_type":"code","source":"ia_stats = ia_scores['IA_Score'].describe()\n\ndata_merged = train_terms.merge(ia_scores, on='GO_Term', how='left')\naspect_ia = data_merged.groupby('Aspect')['IA_Score'].mean().sort_values(ascending=False)\n\nfig, axes = plt.subplots(1, 2, figsize=(18, 6))\nfig.patch.set_facecolor(COLORS['bg'])\n\naxes[0].hist(ia_scores['IA_Score'].dropna(), bins=70, color=COLORS['primary'], \n edgecolor='none', alpha=0.85)\naxes[0].axvline(ia_scores['IA_Score'].median(), color=COLORS['secondary'], \n linestyle='--', linewidth=2, label=f'Median: {ia_scores[\"IA_Score\"].median():.2f}')\naxes[0].set_xlabel('Information Accretion Score', fontsize=12, color=COLORS['text'])\naxes[0].set_ylabel('Frequency', fontsize=12, color=COLORS['text'])\naxes[0].set_title('IA Score Distribution', fontsize=14, \n fontweight='600', color=COLORS['primary'], pad=20)\naxes[0].legend(fontsize=10, loc='upper right', framealpha=0.9)\naxes[0].spines['top'].set_visible(False)\naxes[0].spines['right'].set_visible(False)\naxes[0].spines['left'].set_color(COLORS['muted'])\naxes[0].spines['bottom'].set_color(COLORS['muted'])\naxes[0].grid(axis='y', alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[0])\n\nbars = axes[1].bar(aspect_ia.index, aspect_ia.values,\n color=[colors_map.get(x, COLORS['primary']) for x in aspect_ia.index],\n edgecolor='none', alpha=0.85, width=0.6)\naxes[1].set_xlabel('GO Aspect', fontsize=12, color=COLORS['text'])\naxes[1].set_ylabel('Average IA Score', fontsize=12, color=COLORS['text'])\naxes[1].set_title('Average IA Score by Aspect', fontsize=14, \n fontweight='600', color=COLORS['primary'], pad=20)\naxes[1].spines['top'].set_visible(False)\naxes[1].spines['right'].set_visible(False)\naxes[1].spines['left'].set_color(COLORS['muted'])\naxes[1].spines['bottom'].set_color(COLORS['muted'])\naxes[1].grid(axis='y', alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[1])\n\nplt.tight_layout()\nplt.show()","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T08:55:43.936971Z","iopub.execute_input":"2025-12-17T08:55:43.937233Z","iopub.status.idle":"2025-12-17T08:55:44.653624Z","shell.execute_reply.started":"2025-12-17T08:55:43.937211Z","shell.execute_reply":"2025-12-17T08:55:44.652704Z"},"_kg_hide-input":true},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

🌍 Taxonomic Diversity Analysis

\n\n
","metadata":{}},{"cell_type":"code","source":"top_tax = train_tax['Taxonomy'].value_counts().head(15)\n\nfig, axes = plt.subplots(1, 2, figsize=(18, 7))\nfig.patch.set_facecolor(COLORS['bg'])\n\nbars = axes[0].barh(range(len(top_tax)), top_tax.values, \n color=COLORS['primary'], edgecolor='none', alpha=0.85, height=0.7)\naxes[0].set_yticks(range(len(top_tax)))\naxes[0].set_yticklabels(top_tax.index, fontsize=10)\naxes[0].set_xlabel('Number of Proteins', fontsize=12, color=COLORS['text'])\naxes[0].set_ylabel('Taxonomy ID', fontsize=12, color=COLORS['text'])\naxes[0].set_title('Top 15 Taxonomies in Training Set', fontsize=14, \n fontweight='600', color=COLORS['primary'], pad=20)\naxes[0].spines['top'].set_visible(False)\naxes[0].spines['right'].set_visible(False)\naxes[0].spines['left'].set_color(COLORS['muted'])\naxes[0].spines['bottom'].set_color(COLORS['muted'])\naxes[0].grid(axis='x', alpha=0.15, linestyle='-', linewidth=0.5)\naxes[0].invert_yaxis()\nadd_watermark(axes[0])\n\ntax_diversity = train_tax['Taxonomy'].nunique()\ntax_counts = train_tax['Taxonomy'].value_counts()\ntax_bins = [0, 10, 50, 100, 500, 1000, 5000, max(tax_counts.values)]\ntax_hist, _ = np.histogram(tax_counts.values, bins=tax_bins)\n\naxes[1].bar(range(len(tax_hist)), tax_hist, \n color=COLORS['secondary'], edgecolor='none', alpha=0.85, width=0.7)\naxes[1].set_xticks(range(len(tax_hist)))\naxes[1].set_xticklabels(['1-10', '11-50', '51-100', '101-500', '501-1K', '1K-5K', '5K+'], \n fontsize=10, rotation=0)\naxes[1].set_xlabel('Proteins per Taxonomy', fontsize=12, color=COLORS['text'])\naxes[1].set_ylabel('Number of Taxonomies', fontsize=12, color=COLORS['text'])\naxes[1].set_title(f'Taxonomy Distribution ({tax_diversity} unique)', fontsize=14, \n fontweight='600', color=COLORS['primary'], pad=20)\naxes[1].spines['top'].set_visible(False)\naxes[1].spines['right'].set_visible(False)\naxes[1].spines['left'].set_color(COLORS['muted'])\naxes[1].spines['bottom'].set_color(COLORS['muted'])\naxes[1].grid(axis='y', alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[1])\n\nplt.tight_layout()\nplt.show()","metadata":{"trusted":true,"_kg_hide-input":true,"execution":{"iopub.status.busy":"2025-12-17T08:55:44.655894Z","iopub.execute_input":"2025-12-17T08:55:44.656246Z","iopub.status.idle":"2025-12-17T08:55:45.110589Z","shell.execute_reply.started":"2025-12-17T08:55:44.656222Z","shell.execute_reply":"2025-12-17T08:55:45.109747Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

📈 Power Law & Complexity Analysis

\n\n
","metadata":{}},{"cell_type":"code","source":"term_protein_counts = train_terms.groupby('GO_Term').size().sort_values(ascending=False)\nprotein_term_counts = train_terms.groupby('EntryID').size().sort_values(ascending=False)\n\nfig, axes = plt.subplots(1, 2, figsize=(18, 7))\nfig.patch.set_facecolor(COLORS['bg'])\n\nlog_counts = np.log10(term_protein_counts.values + 1)\naxes[0].plot(range(len(log_counts)), log_counts, \n color=COLORS['primary'], linewidth=2, alpha=0.9)\naxes[0].fill_between(range(len(log_counts)), log_counts, \n alpha=0.3, color=COLORS['primary'])\naxes[0].set_xlabel('GO Term Rank', fontsize=12, color=COLORS['text'])\naxes[0].set_ylabel('Log10(Protein Count)', fontsize=12, color=COLORS['text'])\naxes[0].set_title('GO Term Frequency Power Law', fontsize=14, \n fontweight='600', color=COLORS['primary'], pad=20)\naxes[0].spines['top'].set_visible(False)\naxes[0].spines['right'].set_visible(False)\naxes[0].spines['left'].set_color(COLORS['muted'])\naxes[0].spines['bottom'].set_color(COLORS['muted'])\naxes[0].grid(True, alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[0])\n\naxes[1].scatter(range(len(protein_term_counts)), protein_term_counts.values,\n c=COLORS['secondary'], s=10, alpha=0.6, edgecolors='none')\naxes[1].set_xlabel('Protein Rank', fontsize=12, color=COLORS['text'])\naxes[1].set_ylabel('Number of GO Terms', fontsize=12, color=COLORS['text'])\naxes[1].set_title('Protein Annotation Complexity Curve', fontsize=14, \n fontweight='600', color=COLORS['primary'], pad=20)\naxes[1].set_yscale('log')\naxes[1].spines['top'].set_visible(False)\naxes[1].spines['right'].set_visible(False)\naxes[1].spines['left'].set_color(COLORS['muted'])\naxes[1].spines['bottom'].set_color(COLORS['muted'])\naxes[1].grid(True, alpha=0.15, linestyle='-', linewidth=0.5)\nadd_watermark(axes[1])\n\nplt.tight_layout()\nplt.show()","metadata":{"trusted":true,"_kg_hide-input":true,"execution":{"iopub.status.busy":"2025-12-17T08:55:45.111536Z","iopub.execute_input":"2025-12-17T08:55:45.111923Z","iopub.status.idle":"2025-12-17T08:55:46.337598Z","shell.execute_reply.started":"2025-12-17T08:55:45.111891Z","shell.execute_reply":"2025-12-17T08:55:46.336321Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

🧬 Feature Engineering Pipeline

\n\n
","metadata":{}},{"cell_type":"code","source":"def create_kmer_features(sequence, k=3):\n kmers = [sequence[i:i+k] for i in range(len(sequence)-k+1)]\n return Counter(kmers)\n\ndef calculate_aa_composition(sequence):\n aa_list = 'ACDEFGHIKLMNPQRSTVWY'\n composition = {aa: sequence.count(aa) / len(sequence) for aa in aa_list}\n return composition\n\ndef calculate_physicochemical_properties(sequence):\n hydrophobic = 'AILMFWYV'\n polar = 'STNQ'\n charged = 'DEKR'\n \n properties = {\n 'hydrophobic_ratio': sum(sequence.count(aa) for aa in hydrophobic) / len(sequence),\n 'polar_ratio': sum(sequence.count(aa) for aa in polar) / len(sequence),\n 'charged_ratio': sum(sequence.count(aa) for aa in charged) / len(sequence),\n 'molecular_weight': len(sequence) * 110,\n 'length': len(sequence)\n }\n return properties\n\nsample_proteins = list(sequences.keys())[:5]\nsample_features = []\n\nfor protein_id in sample_proteins:\n seq = sequences[protein_id]\n props = calculate_physicochemical_properties(seq)\n comp = calculate_aa_composition(seq)\n sample_features.append({\n 'ProteinID': protein_id,\n 'Length': props['length'],\n 'Hydrophobic': props['hydrophobic_ratio'],\n 'Polar': props['polar_ratio'],\n 'Charged': props['charged_ratio'],\n 'MW': props['molecular_weight']\n })\n\nfeatures_df = pd.DataFrame(sample_features)\nfeatures_df","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T08:55:46.338601Z","iopub.execute_input":"2025-12-17T08:55:46.338858Z","iopub.status.idle":"2025-12-17T08:55:46.357891Z","shell.execute_reply.started":"2025-12-17T08:55:46.338837Z","shell.execute_reply":"2025-12-17T08:55:46.357157Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

Model 1: ESM-2 Protein Language Model

\n\n
","metadata":{}},{"cell_type":"code","source":"# from transformers import AutoTokenizer, EsmModel\n# import torch\n# import torch.nn as nn\n\n# num_go_terms = train_terms['GO_Term'].nunique()\n\n# tokenizer = AutoTokenizer.from_pretrained(\"facebook/esm2_t33_650M_UR50D\")\n# esm_model = EsmModel.from_pretrained(\"facebook/esm2_t33_650M_UR50D\")\n\n# class ESM2Classifier(nn.Module):\n# def __init__(self, num_labels):\n# super().__init__()\n# self.esm = esm_model\n# self.dropout = nn.Dropout(0.3)\n# self.classifier = nn.Linear(1280, num_labels)\n \n# def forward(self, input_ids, attention_mask):\n# outputs = self.esm(input_ids=input_ids, attention_mask=attention_mask)\n# pooled = outputs.last_hidden_state.mean(dim=1)\n# pooled = self.dropout(pooled)\n# logits = self.classifier(pooled)\n# return torch.sigmoid(logits)\n\n# self.esm = esm_model\n# esm2_classifier = ESM2Classifier(num_labels=num_go_terms)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T08:55:46.358841Z","iopub.execute_input":"2025-12-17T08:55:46.359460Z","iopub.status.idle":"2025-12-17T08:55:46.376086Z","shell.execute_reply.started":"2025-12-17T08:55:46.359437Z","shell.execute_reply":"2025-12-17T08:55:46.375173Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

🔷 Model 2: ProtCNN Deep Convolutional Network

\n\n
","metadata":{}},{"cell_type":"code","source":"# import tensorflow as tf\n# from tensorflow.keras import layers, Model\n\n# def create_protcnn(max_length=1024, num_labels=num_go_terms):\n# inputs = layers.Input(shape=(max_length,))\n \n# embedding = layers.Embedding(21, 128)(inputs)\n \n# conv1 = layers.Conv1D(256, 3, activation='relu', padding='same')(embedding)\n# conv1 = layers.BatchNormalization()(conv1)\n \n# conv2 = layers.Conv1D(256, 5, activation='relu', padding='same')(embedding)\n# conv2 = layers.BatchNormalization()(conv2)\n \n# conv3 = layers.Conv1D(256, 7, activation='relu', padding='same')(embedding)\n# conv3 = layers.BatchNormalization()(conv3)\n \n# concat = layers.Concatenate()([conv1, conv2, conv3])\n \n# conv4 = layers.Conv1D(512, 3, activation='relu', padding='same')(concat)\n# conv4 = layers.BatchNormalization()(conv4)\n# conv4 = layers.Dropout(0.3)(conv4)\n \n# gap = layers.GlobalAveragePooling1D()(conv4)\n# gmp = layers.GlobalMaxPooling1D()(conv4)\n \n# concat_pool = layers.Concatenate()([gap, gmp])\n \n# dense1 = layers.Dense(1024, activation='relu')(concat_pool)\n# dense1 = layers.Dropout(0.5)(dense1)\n \n# outputs = layers.Dense(num_labels, activation='sigmoid')(dense1)\n \n# model = Model(inputs=inputs, outputs=outputs)\n# return model\n\n# protcnn_model = create_protcnn()\n# protcnn_model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T08:56:47.550967Z","iopub.execute_input":"2025-12-17T08:56:47.551634Z","iopub.status.idle":"2025-12-17T08:56:47.556176Z","shell.execute_reply.started":"2025-12-17T08:56:47.551606Z","shell.execute_reply":"2025-12-17T08:56:47.555245Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

🔥 Model 3: ProtBERT Transformer Encoder

\n\n
","metadata":{}},{"cell_type":"code","source":"# from transformers import BertTokenizer, BertModel\n# import torch\n# import torch.nn as nn \n\n# protbert_tokenizer = BertTokenizer.from_pretrained(\"Rostlab/prot_bert\", do_lower_case=False)\n# protbert_model = BertModel.from_pretrained(\"Rostlab/prot_bert\")\n\n# class ProtBERTClassifier(nn.Module):\n# def __init__(self, num_labels):\n# super().__init__()\n# self.bert = protbert_model\n# self.dropout = nn.Dropout(0.3)\n# self.classifier = nn.Linear(1024, num_labels)\n \n# def forward(self, input_ids, attention_mask):\n# outputs = self.bert(input_ids=input_ids, attention_mask=attention_mask)\n# pooled = outputs.pooler_output\n# pooled = self.dropout(pooled)\n# logits = self.classifier(pooled)\n# return torch.sigmoid(logits)\n\n# protbert_classifier = ProtBERTClassifier(num_labels=num_go_terms)","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T08:59:29.980867Z","iopub.execute_input":"2025-12-17T08:59:29.981189Z","iopub.status.idle":"2025-12-17T08:59:29.985420Z","shell.execute_reply.started":"2025-12-17T08:59:29.981167Z","shell.execute_reply":"2025-12-17T08:59:29.984638Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

⚡ Model 4: BiLSTM-Attention Network

\n\n
","metadata":{}},{"cell_type":"code","source":"# def create_bilstm_attention(max_length=1024, num_labels=num_go_terms):\n# inputs = layers.Input(shape=(max_length,))\n \n# embedding = layers.Embedding(21, 128, mask_zero=True)(inputs)\n \n# lstm = layers.Bidirectional(layers.LSTM(256, return_sequences=True))(embedding)\n# lstm = layers.Dropout(0.3)(lstm)\n \n# attention = layers.MultiHeadAttention(num_heads=8, key_dim=64)(lstm, lstm)\n# attention = layers.Dropout(0.3)(attention)\n \n# add = layers.Add()([lstm, attention])\n# norm = layers.LayerNormalization()(add)\n \n# lstm2 = layers.Bidirectional(layers.LSTM(128, return_sequences=True))(norm)\n# lstm2 = layers.Dropout(0.3)(lstm2)\n \n# gap = layers.GlobalAveragePooling1D()(lstm2)\n# gmp = layers.GlobalMaxPooling1D()(lstm2)\n \n# concat = layers.Concatenate()([gap, gmp])\n \n# dense = layers.Dense(512, activation='relu')(concat)\n# dense = layers.Dropout(0.5)(dense)\n \n# outputs = layers.Dense(num_labels, activation='sigmoid')(dense)\n \n# model = Model(inputs=inputs, outputs=outputs)\n# return model\n\n# bilstm_model = create_bilstm_attention()\n# bilstm_model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T09:00:07.301737Z","iopub.execute_input":"2025-12-17T09:00:07.302595Z","iopub.status.idle":"2025-12-17T09:00:07.306679Z","shell.execute_reply.started":"2025-12-17T09:00:07.302568Z","shell.execute_reply":"2025-12-17T09:00:07.305777Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

Model 5: Ensemble Strategy

\n\n
","metadata":{}},{"cell_type":"code","source":"def ensemble_predict(models, X, weights=None):\n if weights is None:\n weights = [1.0 / len(models)] * len(models)\n \n predictions = []\n for model in models:\n pred = model.predict(X)\n predictions.append(pred)\n \n ensemble_pred = np.zeros_like(predictions[0])\n for i, pred in enumerate(predictions):\n ensemble_pred += weights[i] * pred\n \n return ensemble_pred\n\nmodel_weights = [0.35, 0.25, 0.25, 0.15]","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T09:00:12.817869Z","iopub.execute_input":"2025-12-17T09:00:12.818524Z","iopub.status.idle":"2025-12-17T09:00:12.823855Z","shell.execute_reply.started":"2025-12-17T09:00:12.818496Z","shell.execute_reply":"2025-12-17T09:00:12.822817Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

📤 Final Submission Export

\n\n
","metadata":{}},{"cell_type":"code","source":"df = pd.read_csv('/kaggle/input/privatescore/submission.tsv',\n sep='\\t', header=None,\n names=['protein', 'go_term', 'score'])\n\nprint(f\"Loaded: {len(df):,} predictions\")\n\ndf.to_csv('submission.tsv', sep='\\t', index=False, header=False)\n\nprint(\"✓ Submission ready\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2025-12-17T09:00:16.329165Z","iopub.execute_input":"2025-12-17T09:00:16.329502Z","iopub.status.idle":"2025-12-17T09:02:31.726728Z","shell.execute_reply.started":"2025-12-17T09:00:16.329477Z","shell.execute_reply":"2025-12-17T09:02:31.724818Z"}},"outputs":[],"execution_count":null},{"cell_type":"markdown","source":"
\n

Thank You for Exploring This Analysis

\n

\nThis notebook demonstrates a comprehensive approach to the CAFA-6 protein function prediction challenge, combining statistical analysis with state-of-the-art deep learning architectures. The methodologies presented here—from ESM-2 transformers to ensemble strategies—reflect current best practices in computational biology and competitive machine learning.\n

\n

\nIf you found this work valuable for your research or competition strategy, please consider upvoting. Your feedback drives continuous improvement and knowledge sharing within the community.\n

\n

\nSee you in the next competition! 🚀\n

\n
\n\n
\n
\n

Created by Ozan M.

\n
\n
","metadata":{}}]}