b.confidenceScore - a.confidenceScore)
.slice(0, 30)
.map((r) => ({
name: r.name,
score: Math.round(r.confidenceScore * 10000) / 10000,
mechanism: (r.mechanism || "").slice(0, 100),
fda_status: (r.fdaStatus || "").slice(0, 60),
protein: r.proteinTarget || "",
source: r.source || "",
round: r.round || 1,
}))
const system =
"You are a computational pharmacology researcher analyzing molecular " +
"docking results for drug repurposing. You must return ONLY valid JSON.\n\n" +
"Analyze the top docking hits and look for patterns:\n" +
"- Are drugs from the same therapeutic class clustering as top hits?\n" +
" (e.g., multiple statins, multiple SSRIs, multiple kinase inhibitors)\n" +
"- Do top-scoring compounds share structural features?\n" +
"- Are there unexplored drug classes that might work?\n\n" +
"Return a JSON object with:\n" +
' "action": one of "expand_3d_similar", "expand_class", or "proceed"\n' +
' "rationale": your analysis of the patterns (2-3 sentences)\n' +
' "hypothesis": a specific scientific hypothesis (1 sentence)\n' +
' "seed_cids": [list of CID numbers] (if action is expand_3d_similar)\n' +
' "drug_class": "class name" (if action is expand_class)\n' +
' "drug_names": ["drug1", "drug2", ...] (if action is expand_class)\n\n' +
"Rules:\n" +
`- This is round ${round}. Max rounds: ${MAX_EXPANSION_ROUNDS + 1}.\n` +
`- If round >= ${MAX_EXPANSION_ROUNDS + 1}, you MUST choose "proceed".\n` +
'- Only choose "expand" if you see a clear pattern worth investigating.\n' +
"- If expanding, be specific about which CIDs or drug names to test."
const user =
`Cancer type: ${cancerType}\n` +
`Round: ${round}\n` +
`Previous hypotheses: ${JSON.stringify(hypotheses)}\n` +
`Previous expansions: ${JSON.stringify(expansionHistory)}\n\n` +
`Top 30 docking results (sorted by confidence score):\n` +
`${JSON.stringify(topResults, null, 2)}\n\n` +
`Analyze these results and return your decision as JSON.`
const raw = await queryPerplexity(system, user)
const parsed = parseJsonResponse(raw) as Record
const decision: ReasoningDecision = {
action: "proceed",
rationale: (parsed.rationale as string) || "",
hypothesis: (parsed.hypothesis as string) || "",
}
// Validate action
const action = parsed.action as string
if (["expand_3d_similar", "expand_class", "proceed"].includes(action)) {
decision.action = action as ReasoningDecision["action"]
}
// Force proceed if max rounds reached
if (round >= MAX_EXPANSION_ROUNDS + 1) {
decision.action = "proceed"
decision.rationale += " (Max rounds reached, proceeding to synthesis.)"
}
if (decision.action === "expand_3d_similar") {
decision.seedCids = (parsed.seed_cids as number[]) || []
}
if (decision.action === "expand_class") {
decision.drugClass = (parsed.drug_class as string) || ""
decision.drugNames = (parsed.drug_names as string[]) || []
}
return NextResponse.json(decision)
} catch (error) {
console.error("Analyze API error:", error)
return NextResponse.json(
{ error: error instanceof Error ? error.message : "Unknown error" },
{ status: 500 }
)
}
}
```
### app/api/report/route.ts
```typescript
import { NextResponse } from "next/server"
import { queryPerplexity } from "@/lib/perplexity"
import type { DockingResult, DockingTarget } from "@/lib/types"
export const maxDuration = 180
function classifyCompound(
fdaStatus: string,
cancerType: string
): "cancer_purposed" | "repurposing" | "novel" {
const s = fdaStatus.toLowerCase()
const cancer = cancerType.toLowerCase()
if (s.includes("approved") && s.includes(cancer.split(/\s+/)[0])) {
return "cancer_purposed"
}
if (s.includes("approved") || s.includes("fda")) {
return "repurposing"
}
return "novel"
}
export async function POST(req: Request) {
try {
const { cancerType, allResults, targets } = (await req.json()) as {
cancerType: string
allResults: DockingResult[]
targets: DockingTarget[]
}
if (!cancerType || !allResults?.length) {
return NextResponse.json(
{ error: "cancerType and allResults are required" },
{ status: 400 }
)
}
// Build overview per target
const resultsByProtein: Record = {}
for (const r of allResults) {
const prot = r.proteinTarget || "unknown"
if (!resultsByProtein[prot]) resultsByProtein[prot] = []
resultsByProtein[prot].push(r)
}
// Sort each target's results
for (const prot of Object.keys(resultsByProtein)) {
resultsByProtein[prot].sort(
(a, b) => b.confidenceScore - a.confidenceScore
)
}
// Classify compounds
const classified: Record = {
cancer_purposed: [],
repurposing: [],
novel: [],
}
for (const r of allResults) {
const cat = classifyCompound(r.fdaStatus, cancerType)
classified[cat].push(r)
}
// Build overview text
let overviewText = `## Docking Overview\n\n`
overviewText += `- Cancer type: ${cancerType}\n`
overviewText += `- Total compounds docked: ${allResults.length}\n`
overviewText += `- Protein targets: ${targets.map((t) => t.protein).join(", ")}\n\n`
for (const target of targets) {
const results = resultsByProtein[target.protein] || []
overviewText += `### ${target.protein} (PDB: ${target.pdbId || "N/A"})\n\n`
overviewText += `| Rank | Drug | Score | FDA Status | Mechanism |\n`
overviewText += `|------|------|-------|------------|------------|\n`
for (const [i, r] of results.slice(0, 15).entries()) {
overviewText += `| ${i + 1} | ${r.name} | ${r.confidenceScore.toFixed(4)} | ${r.fdaStatus} | ${r.mechanism.slice(0, 60)} |\n`
}
overviewText += "\n"
}
const classifiedText =
`Cancer-purposed drugs: ${classified.cancer_purposed.length}\n` +
`Repurposing candidates: ${classified.repurposing.length}\n` +
`Novel/research compounds: ${classified.novel.length}`
// Generate methodology via Perplexity
const methodologyPrompt =
`Write a detailed methodology section for a drug repurposing paper about ${cancerType}. ` +
`The study used:\n` +
`- DiffDock (diffusion-based molecular docking model) on RunPod serverless GPU\n` +
`- 10 poses per ligand-protein pair\n` +
`- Protein structures from RCSB PDB\n` +
`- Ligands from PubChem + curated databases\n` +
`- Perplexity AI for literature analysis and reasoning\n\n` +
`Overview:\n${overviewText}`
const methodology = await queryPerplexity(
"You are a scientific writer. Write a detailed methodology section in Markdown.",
methodologyPrompt
)
// Generate results via Perplexity
const resultsPrompt =
`Write a detailed results section for a drug repurposing paper about ${cancerType}.\n\n` +
`Docking results:\n${overviewText}\n\n` +
`Classification:\n${classifiedText}\n\n` +
`Focus on:\n` +
`1. Overview of docking runs\n` +
`2. Drug repurposing leaderboard (ALL FDA drugs not approved for this cancer)\n` +
`3. Novel/research compounds\n` +
`4. Cross-target comparison\n` +
`5. Score distribution analysis`
const results = await queryPerplexity(
"You are a scientific writer. Write a detailed results section in Markdown with tables.",
resultsPrompt
)
// Generate conclusion via Perplexity
const conclusionPrompt =
`Write a conclusion for a drug repurposing paper about ${cancerType}.\n\n` +
`Results:\n${overviewText}\n\n` +
`Classification:\n${classifiedText}\n\n` +
`Cover: key findings, top repurposing candidates, clinical implications, limitations, future directions.`
const conclusion = await queryPerplexity(
"You are a scientific writer. Write a conclusion section in Markdown.",
conclusionPrompt
)
const resultsMd =
`# Results: Drug Repurposing Analysis for ${cancerType}\n\n` +
`## Methodology\n\n${methodology}\n\n` +
`## Results\n\n${results}\n\n` +
`## Conclusion\n\n${conclusion}\n`
return NextResponse.json({ resultsMd })
} catch (error) {
console.error("Report API error:", error)
return NextResponse.json(
{ error: error instanceof Error ? error.message : "Unknown error" },
{ status: 500 }
)
}
}
```
[87 more indexed source files omitted to keep this export small. The full file list is in the Codebase structure section above.]