Spaces:
Build error
Build error
File size: 18,364 Bytes
9d976fc | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 | """
decision_engine.py
==================
Unified decision logic β probability, dominant mechanism, confidence level,
and human-readable interpretation grounded in internal signals.
Prediction is NEVER computed without the explainability analysis being
present first. This module enforces that ordering.
"""
from __future__ import annotations
import json
from dataclasses import dataclass, field
from typing import Optional
import numpy as np
from explainability_engine import (
SpliceSignals,
V4Signals,
ClassicSignals,
_classify_risk_tier,
)
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Result dataclass
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@dataclass
class DecisionResult:
variant: str
unified_probability: float
risk_tier: str
tier_desc: str
dominant_mechanism: str
confidence: str
# per-model
splice_prob: float
v4_prob: float
classic_prob: float
# XAI engine metrics
mutation_peak_ratio: float
counterfactual_magnitude:float
cross_model_locality: float
signal_concentration: float
explainability_strength: float
activation_pattern: str
# interpretation
splice_analysis: str
protein_analysis: str
agreement_analysis: str
final_explanation: str
# full structured JSON (as string)
report_json: str = field(repr=False)
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Dominant mechanism logic
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _dominant_mechanism(splice: SpliceSignals, v4: V4Signals,
classic: ClassicSignals,
prob_std: float) -> str:
"""
Determine which mechanism drives the prediction.
Splice-driven: splice model dominates AND aura / splice importance elevated
Protein-driven: v4 + classic > splice probability by meaningful margin
Consensus: all three models agree within 0.10
Ambiguous: high disagreement (prob_std > 0.12)
"""
probs = np.array([splice.probability, v4.probability, classic.probability])
p_max = float(probs.max())
p_min = float(probs.min())
if prob_std > 0.14:
return "Ambiguous"
# Splice dominance
splice_leads = splice.probability > v4.probability + 0.05
high_aura = splice.splice_aura_score > 0.35
high_splice_i = float(splice.splice_imp.max()) > 0.50
if splice_leads and (high_aura or high_splice_i):
return "Splice-driven"
# Protein dominance
protein_avg = (v4.probability + classic.probability) / 2
if protein_avg > splice.probability + 0.05 and float(splice.region_imp[0]) > 0.5:
return "Protein-driven"
# Consensus
if p_max - p_min <= 0.10:
return "Consensus"
return "Ambiguous"
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Confidence level
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _confidence(unified_prob: float, prob_std: float,
ess: float, cf_mag: float) -> str:
"""
High: strong signal from all three axes
Moderate: partial support
Low: conflicting or weak signals
"""
score = 0
# Model agreement
if prob_std < 0.05: score += 2
elif prob_std < 0.12: score += 1
# Explainability strength
if ess >= 0.65: score += 2
elif ess >= 0.35: score += 1
# Counterfactual magnitude
if cf_mag >= 0.25: score += 2
elif cf_mag >= 0.10: score += 1
# Probability extremity
if unified_prob >= 0.85 or unified_prob <= 0.15: score += 1
if score >= 5: return "High"
if score >= 3: return "Moderate"
return "Low"
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Human-readable interpretation builders
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def _splice_analysis_text(s: SpliceSignals, variant: str) -> str:
parts = []
tier, _ = _classify_risk_tier(s.probability)
parts.append(
f"The splice model classifies variant {variant} as "
f"'{tier}' with probability {s.probability:.4f}."
)
if s.mutation_peak_ratio >= 2.0:
parts.append(
f"The conv3 activation peak at the mutation site "
f"(position {s.mutation_pos}) is {s.mutation_peak_ratio:.2f}Γ "
f"the mean window activation, indicating the model is strongly "
f"attending to the mutation location."
)
elif s.mutation_peak_ratio >= 1.0:
parts.append(
f"The mutation position ({s.mutation_pos}) has above-average "
f"conv3 activation (MPR={s.mutation_peak_ratio:.2f}Γ)."
)
else:
parts.append(
f"The mutation position ({s.mutation_pos}) does not carry an "
f"elevated conv3 activation peak (MPR={s.mutation_peak_ratio:.2f}Γ), "
f"suggesting the pathogenic signal may arise from broader context."
)
if s.splice_risk_donor in ("CRITICAL SPLICE SITE", "SPLICE REGION"):
parts.append(
f"The variant lies {s.dist_donor} bp from the nearest GT donor "
f"dinucleotide (risk: {s.splice_risk_donor}). "
f"Splice donor importance score = {float(s.splice_imp[0]):.3f}."
)
if s.splice_risk_acceptor in ("CRITICAL SPLICE SITE", "SPLICE REGION"):
parts.append(
f"The variant lies {s.dist_acceptor} bp from the nearest AG acceptor "
f"dinucleotide (risk: {s.splice_risk_acceptor}). "
f"Splice acceptor importance score = {float(s.splice_imp[1]):.3f}."
)
cf = s.counterfactual
if cf.get("probability_range", 0) > 0.15:
parts.append(
f"Counterfactual analysis: swapping the alternate base changes "
f"the pathogenicity probability by up to "
f"{cf['probability_range']:.3f} "
f"(range {cf['min_probability']:.3f}β{cf['max_probability']:.3f}), "
f"confirming strong position-level causality."
)
abl = s.ablation
dom = abl.get("dominant_feature", "unknown")
parts.append(
f"Feature ablation: '{dom}' contributes the largest share "
f"of the pathogenic signal "
f"({abl.get(dom.lower().split()[0]+'_pct', '?')}%)."
)
return " ".join(parts)
def _protein_analysis_text(v4: V4Signals, classic: ClassicSignals) -> str:
parts = []
v4_tier, _ = _classify_risk_tier(v4.probability)
cl_tier, _ = _classify_risk_tier(classic.probability)
parts.append(
f"V4 model: '{v4_tier}' (prob={v4.probability:.4f}). "
f"Classic model: '{cl_tier}' (prob={classic.probability:.4f})."
)
avg_mpr = (v4.mutation_peak_ratio + classic.mutation_peak_ratio) / 2
parts.append(
f"Average mutation-site activation ratio across protein models: "
f"{avg_mpr:.2f}Γ."
)
if float(classic.region_imp[0]) > 0.6:
parts.append(
"The classic model assigns high exon-region importance "
f"({float(classic.region_imp[0]):.3f}), "
"supporting a protein-coding disruption mechanism."
)
elif float(classic.region_imp[1]) > 0.6:
parts.append(
"The classic model assigns high intron-region importance "
f"({float(classic.region_imp[1]):.3f}), "
"consistent with a regulatory or splicing mechanism."
)
return " ".join(parts)
def _agreement_text(splice: SpliceSignals, v4: V4Signals,
classic: ClassicSignals, cross_locality: float,
prob_std: float) -> str:
probs = [splice.probability, v4.probability, classic.probability]
p_max = max(probs)
p_min = min(probs)
lvl = "agree" if prob_std < 0.10 else ("partially agree" if prob_std < 0.18 else "disagree")
lines = [
f"The three models {lvl} on pathogenicity "
f"(probabilities: splice={splice.probability:.3f}, "
f"v4={v4.probability:.3f}, classic={classic.probability:.3f}; "
f"std={prob_std:.3f})."
]
lines.append(
f"Cross-model activation locality score = {cross_locality:.4f} "
f"({'high' if cross_locality > 0.5 else 'moderate' if cross_locality > 0.0 else 'low'} "
f"alignment of importance peaks across models)."
)
if cross_locality > 0.5:
lines.append(
"All three models are attending to the same region of the 99-bp window, "
"strongly supporting the identified mechanistic signal."
)
else:
lines.append(
"The models are attending to different sequence regions, "
"suggesting the signal may be mechanism-specific or context-dependent."
)
return " ".join(lines)
def _final_explanation(result: "DecisionResult") -> str:
prob_label = (
"highly pathogenic" if result.unified_probability >= 0.85 else
"likely pathogenic" if result.unified_probability >= 0.70 else
"possibly pathogenic" if result.unified_probability >= 0.50 else
"likely benign" if result.unified_probability >= 0.20 else
"benign"
)
mech = result.dominant_mechanism.lower()
ess = result.explainability_strength
lines = [
f"Variant {result.variant} is predicted {prob_label} "
f"(unified probability = {result.unified_probability:.4f}; "
f"risk tier: {result.risk_tier}) with {result.confidence.lower()} confidence.",
f"The prediction is driven by a {mech} mechanism.",
f"Explainability strength score = {result.explainability_strength:.4f} "
f"({'high' if ess >= 0.65 else 'moderate' if ess >= 0.35 else 'low'} "
f"overall evidence quality), based on a mutation peak ratio of "
f"{result.mutation_peak_ratio:.2f}Γ, counterfactual magnitude of "
f"{result.counterfactual_magnitude:.4f}, and cross-model locality of "
f"{result.cross_model_locality:.4f}.",
f"The conv3 activation pattern is classified as '{result.activation_pattern}', "
f"and signal concentration at the mutation site is "
f"{result.signal_concentration:.4f}.",
f"Model agreement is {result.agreement_analysis.lower()}.",
]
lines.append(
"β This system is for research use only and should not substitute "
"for clinical diagnostic testing."
)
return "\n\n".join(lines)
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Main entry point
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def build_decision(chrom: str, pos: int, ref: str, alt: str,
splice: SpliceSignals,
v4: V4Signals,
classic: ClassicSignals,
cross: dict) -> DecisionResult:
"""
Build the unified DecisionResult. Called AFTER all XAI signals have been
extracted β this ordering is mandatory.
"""
variant = f"chr{chrom}:{pos} {ref}>{alt}"
# Weighted average probability β splice weighted 0.45 (most informative),
# v4 0.30, classic 0.25
unified = (0.45 * splice.probability +
0.30 * v4.probability +
0.25 * classic.probability)
unified = round(float(unified), 4)
tier, tier_desc = _classify_risk_tier(unified)
prob_std = cross["prob_std"]
mech = _dominant_mechanism(splice, v4, classic, prob_std)
ess = cross["explainability_strength_score"]
cf_mag = cross["counterfactual_magnitude"]
conf = _confidence(unified, prob_std, ess, cf_mag)
splice_txt = _splice_analysis_text(splice, variant)
protein_txt = _protein_analysis_text(v4, classic)
agreement_txt= _agreement_text(splice, v4, classic,
cross["cross_model_locality_score"], prob_std)
# Build result skeleton so _final_explanation can reference it
r = DecisionResult(
variant=variant,
unified_probability=unified,
risk_tier=tier, tier_desc=tier_desc,
dominant_mechanism=mech, confidence=conf,
splice_prob=splice.probability,
v4_prob=v4.probability,
classic_prob=classic.probability,
mutation_peak_ratio=cross["mutation_peak_ratio"],
counterfactual_magnitude=cross["counterfactual_magnitude"],
cross_model_locality=cross["cross_model_locality_score"],
signal_concentration=cross["signal_concentration_index"],
explainability_strength=ess,
activation_pattern=cross["activation_pattern_type"],
splice_analysis=splice_txt,
protein_analysis=protein_txt,
agreement_analysis=agreement_txt,
final_explanation="", # filled below
report_json="", # filled below
)
r.final_explanation = _final_explanation(r)
# ββ Build structured JSON βββββββββββββββββββββββββββββββββββββββββββββββββ
cf = splice.counterfactual
abl= splice.ablation
report = {
"variant": variant,
"prediction": {
"unified_probability": unified,
"risk_tier": tier,
"tier_desc": tier_desc,
"dominant_mechanism": mech,
"confidence": conf,
},
"model_outputs": {
"splice": {
"probability": splice.probability,
"risk_tier": splice.risk_tier,
"conv3_peak_at_mutation": (
round(float(splice.conv3_norm[splice.mutation_pos]), 4)
if 0 <= splice.mutation_pos < 99 else None
),
"splice_importance": {
"donor": round(float(splice.splice_imp[0]), 4),
"acceptor":round(float(splice.splice_imp[1]), 4),
"region": round(float(splice.splice_imp[2]), 4),
},
"region_importance": {
"exon": round(float(splice.region_imp[0]), 4),
"intron": round(float(splice.region_imp[1]), 4),
},
"splice_aura_score": splice.splice_aura_score,
"dist_donor": splice.dist_donor,
"dist_acceptor": splice.dist_acceptor,
"splice_risk_donor": splice.splice_risk_donor,
"splice_risk_acceptor": splice.splice_risk_acceptor,
"counterfactual": {k: v for k, v in cf.items()
if k != "table"},
"counterfactual_table": cf.get("table", []),
"feature_ablation": abl,
},
"v4": {
"probability": v4.probability,
"mutation_peak_ratio": round(v4.mutation_peak_ratio, 4),
"signal_concentration": round(v4.signal_concentration, 4),
},
"classic": {
"probability": classic.probability,
"importance_head": round(classic.importance_head, 4),
"region_importance": {
"exon": round(float(classic.region_imp[0]), 4),
"intron": round(float(classic.region_imp[1]), 4),
},
"mutation_peak_ratio": round(classic.mutation_peak_ratio, 4),
},
},
"explainability_analysis": {
"mutation_peak_ratio": cross["mutation_peak_ratio"],
"counterfactual_magnitude": cross["counterfactual_magnitude"],
"cross_model_locality_score": cross["cross_model_locality_score"],
"signal_concentration_index": cross["signal_concentration_index"],
"explainability_strength_score": ess,
"activation_pattern_type": cross["activation_pattern_type"],
"model_agreement": cross["model_agreement"],
"prob_std": cross["prob_std"],
},
"interpretation": {
"splice_analysis": splice_txt,
"protein_analysis": protein_txt,
"agreement_analysis":agreement_txt,
"final_explanation": r.final_explanation,
},
}
r.report_json = json.dumps(report, indent=2, default=str)
return r
|