BD Match Pro

Where does your asset belong?

你的资产，该卖给谁？

AI · China Assets · Global Markets AI · 中国资产 · 全球市场

🌐 Try the live tool · 立即体验 · 🐙 Source · 🗺️ Roadmap · 💬 Discussions

---

🇨🇳 中文

BD Match Pro 是一个开源的、AI 驱动的跨境医药授权交易匹配与估值引擎。它针对医药 BD 工作的"第一公里"——这个品种应该卖给谁？值多少钱？怎么个卖法？——提供数据化、参数化、可解释的初步答案。

它能做什么

1. 匹配买家——基于 75 家全球战略买家的精选画像（组合策略、地理偏好、近期交易动作），对一个具体品种打分排序，给出最契合的对手方清单 + 公开联系方式。
2. 估值与交易结构——把 rNPV 风险调整模型 与 可比交易回归 按 50/50 权重融合(可比交易库当前 328 笔，覆盖 2024-11 → 2026-02 公开披露的跨境授权交易；其中 ~50 笔为人工精选含完整临床阶段，其余为批量摄入、临床阶段标记 unknown、药物类型按药名规则推断)，给出首付款、里程碑、销售提成的合理区间，并推荐 License-out / NewCo / 联合开发 / 区域分割 等交易结构。当匹配样本 < 3 时模型自动退化为纯 rNPV，并在结果区显式提示。
3. 参数全开放——成功率（PoS）、贴现率（WACC）、峰值销售、毛利率、爬坡曲线、专利期等核心参数全部可由用户自由调整，实时看到不同假设下的估值变化。
4. 方法学透明——所有公式、参数默认值、买方权重逻辑、相似度加权方式均在源代码与方法学面板中公开，欢迎挑战、修改、改进。

关于数据库规模与质量的诚实说明：v1.0.2 库共 328 笔。其中 53 笔为 v1.0 起就已经手工精选录入的高质量条目（含完整临床阶段、备注），其余 275 笔为从公开披露数据批量摄入——临床阶段一律标记为 unknown，药物类型 (modality) 通过药名规则推断（如 -mab → 单抗、-tinib → 小分子、ADC / CAR-T 通过显式 token）。短期路线：(a) 扩充至 500+，(b) 给批量条目回填临床阶段。欢迎在 Issues 提交补充数据。

为什么开源

医药 BD 长期是少数大行与精英咨询的"黑箱"。本项目想做一件简单的事：把第一步的方法论开放出来，让中国生物医药资产更高效地走向全球市场。

交易是艺术，不是模型。 真实成交逻辑与任何预测都可能大相径庭。但只要在 BD 的第一公里，AI 能让大家少走点弯路、提升一点效率，就已经有用了。

---

🇬🇧 English

BD Match Pro is an open-source, AI-powered cross-border licensing engine for pharmaceutical assets. It tackles the first mile of biotech business development — who would buy this asset, at what price, and via which deal structure? — with data-driven, fully parametric, and transparent first-pass answers.

What it does

1. Match buyers — score and rank 75 deeply profiled global strategic acquirers against your asset, based on portfolio fit, geographic preference and recent deal activity. Output includes buyer profile, financial snapshot, recent transactions and public contact details.
2. Value & structure — blend risk-adjusted NPV with a comparable-deal regression at 50/50 weight (the comp database currently holds 328 publicly disclosed cross-border deals from 2024-11 → 2026-02; ~50 are hand-curated with full clinical-stage detail, the remainder are bulk-ingested with clinical stage marked unknown and modality inferred from drug-name patterns); size upfront / milestones / royalty ranges; recommend License-out, NewCo, Co-development or Regional split structures. When fewer than 3 similar comps exist, the model falls back to pure rNPV — transparently flagged in the result panel.
3. Fully parametric — every input (PoS, WACC, peak sales, margin, ramp curve, patent runway, etc.) is user-adjustable and the valuation re-renders in real time.
4. Transparent methodology — every formula, default parameter, weight and similarity-scoring rule is documented in the methodology panel and in the source. Challenge it, fork it, improve it.

An honest note about database size and quality: the v1.0.2 database is 328 deals total — 53 hand-curated entries from v1.0 (full clinical-stage detail and editorial notes), plus 275 bulk-ingested from public deal data with clinical stage marked unknown and modality inferred from drug-name patterns (e.g. -mab → mAb, -tinib → small molecule, ADC / CAR-T from explicit tokens). Near-term priorities: (a) expand to 500+, (b) backfill clinical stage on the bulk-ingested entries. Submit additional public deals via Issues.

Why open-source

Pharma BD has long been the black box of a handful of bulge-bracket banks and elite consultancies. This project does one simple thing: open up the first-pass methodology so that Chinese biopharma assets can find their way to global markets more efficiently.

Deals are art, not models. Real-world outcomes diverge from any forecast. But if AI can save practitioners a few wrong turns at the very first mile, it has already earned its keep.

---

🚀 Quick start · 快速开始

Run locally · 本地运行

git clone https://github.com/raphahealthtech/bd-match-pro.git
cd bd-match-pro
# Just open index.html in any modern browser
open index.html              # macOS
xdg-open index.html          # Linux
start index.html             # Windows

The entire app is a single self-contained index.html file — no build step, no backend, no dependencies. All logic lives in plain JavaScript modules within the file. 整个应用是单一的 index.html 文件——零构建、零后端、零依赖，所有逻辑都用纯 JavaScript 写在文件里。

Live demo · 在线体验

→ https://raphahealthtech.github.io/bd-match-pro/

---

🧮 Methodology · 方法学

The valuation engine runs two independent paths and blends them at 50/50 (when ≥ 3 similar comps exist):

1 · Parametric path — Risk-adjusted NPV (rNPV)

rNPV_licensor = PoS × Σ_t  [ Peak × Ramp(t) × ModPremium × CompMult × RoyaltyRate ] / (1 + WACC)^t
rNPV_asset    = PoS × Σ_t  [ Peak × Ramp(t) × ModPremium × CompMult × NetMargin ]   / (1 + WACC)^t

• PoS_cumulative — Preclinical 5% → Phase I 10% → Phase II 17% → Phase III 52% → NDA 88% → Approved 100% (BIO/QLS/Informa benchmarks, user-editable)
• WACC — default 11% (user-adjustable)
• Patent runway — 10 years post-approval; ramp curve 10/30/60/85/100/100/95/80/55/25
• ModPremium — sm 1.00 · mab 1.05 · bsab 1.25 · adc 1.40 · cart 1.50 · gt 1.55 · pep 1.10 · olig 1.20 · rc 1.45

2 · Data path — Comparable-deal regression

For each deal in DEALS_DB (currently 328 cross-border licensing deals with disclosed economics — 53 hand-curated with full detail, 275 bulk-ingested with stage:unknown and inferred modality), compute a similarity weight:

weight(d) = 0.02 (baseline)
          + 0.50 (indication direct hit) | 0.25 (adjacent) | 0.10 (same TA)
          + 0.30 (modality direct hit)   | 0.08 × ratio(ModPremium) (mismatch)
          + 0.20 (stage hit) | 0.10 (±1 stage) | 0.04 (±2 stages)

Deals passing a real-similarity gate (indication ≥ adjacent OR modality direct match OR weak triple of indication/modality/stage) are weighted-resampled, then quantiles taken:

total_implied   = median(samples) × stage_correction × mod_correction × terr_factor
upfront_implied = median(samples.upfront)         × (same correction terms)
P10 / P90       = 20% / 80% quantiles of resampled distribution

3 · Blended output

total_blended    = 0.5 × rNPV_total   + 0.5 × total_implied      // when N ≥ 3
upfront_blended  = 0.5 × rNPV_upfront + 0.5 × upfront_implied
P10 / P90        = union(rNPV sensitivity, comp quantile)        // conservative

When fewer than 3 similar comps exist, the model falls back to pure rNPV and the result panel explicitly says so.

4 · Biosimilar valuation

A separate model with biosim-specific PoS (higher), WACC (lower, 9%), gross margin (~55%), share-of-originator ramp curve, and discount-vs-originator parameter. Biosim does not currently use comp-blending — DEALS_DB carries too few biosim transactions to support the regression.

5 · Buyer matching

Score = 40·IndicationFit + 25·ModalityFit + 20·StageFit + 10·FinancialCapacity + 5·RecentActivity

Each axis is documented in the methodology panel with its sub-weights.

Known limitations

• Modest sample — 328 disclosed-economics deals; statistical confidence improves with sample size but P10/P90 bands still run wide for narrow indication × stage × modality slices.
• Inferred fields on 275 of 328 entries — clinical stage is unknown and modality is inferred from drug-name patterns; an entry with ind:unknown contributes only via modality match in the similarity-weighted regression and is appropriately downweighted in the median, but it does inflate the displayed N.
• Selection bias — database over-indexes China license-outs vs. pure overseas in-licensing.
• Biobucks inflation — reported "total deal value" often inflates distant sales milestones.
• No biosim regression — too few biosim deals to support a separate regression.

Full methodology panel is in the running app (Methodology tab); all parameters are visible & editable in the source index.html. 完整方法学在运行中应用的方法学面板里，所有参数都在源码 index.html 中可见、可改。

---

🤝 Contributing · 参与贡献

We welcome: 欢迎以下贡献：

• 📊 Better default parameters — newer industry data on PoS, WACC, peak sales, ramp curves 更准的默认参数——更新的行业数据（PoS / WACC / 峰值 / 爬坡）
• 💼 Buyer database updates — new pipeline portfolios, recent deals, contact corrections 买方数据库更新——新管线组合、近期交易、联系方式修正
• 🧠 Methodology improvements — better risk-adjustment math, biosim modelling, NewCo structures 方法学改进——更好的风险调整、生物类似药建模、NewCo 结构
• 🌍 Indication / TA expansion — finer subdivisions, regional adjustments 适应证 / 治疗领域扩充——更细分类、地域调整
• 🎨 Design & UX — accessibility, mobile, internationalisation 设计与体验——可达性、移动端、国际化
• 🐛 Bug reports & corrections Bug 反馈与修正

How to contribute · 如何参与

1. Open an Issue to discuss the change first (especially for methodology changes). 先开 Issue 讨论（尤其是方法学改动）。
2. Fork → branch → PR. Keep PRs focused; explain why in the description. Fork → 开分支 → 提 PR；保持改动聚焦，描述里说清为什么。
3. Comment thresholds in the code if you change a default constant — readers need to know what changed and why. 修改默认常量请在代码里加注释，让读者知道改了什么、为什么。

---

📜 License · 许可

Released under the MIT License. You may use, modify, distribute, and even commercialise this software freely, provided you retain the copyright notice. 基于 MIT 协议 发布。任何人都可以自由使用、修改、分发，甚至商用，只需保留版权声明。

⚠️ Not investment advice. Outputs are research-grade estimates. See DISCLAIMER for the full notice. ⚠️ 不构成投资建议。 输出仅为研究级估算，完整免责声明见 DISCLAIMER。

---

💬 Contact · 联系我们

Rapha Health Investment Technology Limited · 睿孚健康投资科技有限公司 Hong Kong · 香港

• 📞 Mainland China · 中国大陆: +86 13632302814
• 📞 Hong Kong · 香港: +852 96602635
• 🌐 Live tool: raphahealthtech.github.io/bd-match-pro
• 💡 Feedback / Collaboration / Custom builds: open an Issue or start a Discussion

---

From China to the World — the first mile, in the open. 中国资产 · 全球市场——第一公里，公开可见。