BAAR Logo

BAAR3-drug-dti

๐ŸŒ Official Website: https://baar.uk/model/
๐Ÿ“„ License: Proprietary Paid License (Paid License Required for Both Research & Commercial Use)


๐Ÿ“Œ Model Overview (๋ชจ๋ธ ๊ฐœ์š”)

BAAR3-drug-dti is an ultra-high-throughput, edge-deployable AI drug discovery engine engineered for real-time Drug-Target Interaction (DTI) and binding affinity ($pK_d$) prediction. Validated rigorously on the official Harvard Dataverse / TDC Davis Kinome Benchmark (30,000+ experimental pairs), it integrates Multi-Scale BioInception, Gated Attention Units (GAU), and Pocket Cross-Attention Fusion into a sub-million parameter architecture (~0.90M params). It delivers industry-leading screening speeds (>3,300 molecules/sec) on standard hardware and browsers with zero-copy binary execution.

BAAR3-drug-dti๋Š” ํ•˜๋ฒ„๋“œ ๋Œ€ํ•™๊ต TDC Davis Kinome ๊ณต์‹ ๋ฒค์น˜๋งˆํฌ(30,000+ ์‹ค์ œ ์‹คํ—˜์Œ)๋กœ ๊ฒ€์ฆ๋œ ์ดˆ๊ณ ์† ์—ฃ์ง€ ๊ตฌ๋™ํ˜• ์•ฝ๋ฌผ-ํ‘œ์  ์ƒํ˜ธ์ž‘์šฉ(DTI) ๋ฐ ๊ฒฐํ•ฉ ์นœํ™”๋„($pK_d$) ์˜ˆ์ธก AI ์—”์ง„์ž…๋‹ˆ๋‹ค. ๊ฑฐ๋Œ€ Bio-LLM ์ˆ˜์ค€์˜ ์žฅ๊ฑฐ๋ฆฌ ๊ฒฐํ•ฉ ์ƒํ˜ธ์ž‘์šฉ ํฌ์ฐฉ ๋Šฅ๋ ฅ์„ 0.9M ์ดˆ๊ฒฝ๋Ÿ‰ ํŒŒ๋ผ๋ฏธํ„ฐ์— ์ง‘์•ฝํ•˜์—ฌ, ๊ธฐ์กด ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ๋ฐ˜ ์‹ ์•ฝ ๋ชจ๋ธ ๋Œ€๋น„ **์ตœ๋Œ€ 300๋ฐฐ ๋น ๋ฅธ ์†๋„(์ดˆ๋‹น 3,300๊ฐœ ์ด์ƒ ๋ถ„์ž ์Šคํฌ๋ฆฌ๋‹)**๋กœ ์œ ํšจ ๊ฒฐํ•ฉ ํ›„๋ณด ๋ฌผ์งˆ์„ ์‹ค์‹œ๊ฐ„ ์„ ๋ณ„ํ•ฉ๋‹ˆ๋‹ค.

๐Ÿงฌ Core Screening Capabilities (์ฃผ์š” ํ•ต์‹ฌ ์—ญ๋Ÿ‰)

  • ์ •๋ฐ€ ๊ฒฐํ•ฉ ์นœํ™”๋„ ์˜ˆ์ธก ($pK_d$): ์ €๋ถ„์ž ํ™”ํ•ฉ๋ฌผ(SMILES)๊ณผ ํ‘œ์  ํ‚ค๋‚˜์•„์ œ(Amino Acid) ์„œ์—ด ๊ฐ„ ๊ฒฐํ•ฉ๋ ฅ($pK_d = -\log_{10} K_d$)์„ ์†Œ์ˆ˜์  ๋‹จ์œ„๋กœ ์ •๋ฐ€ ํšŒ๊ท€ ์ถ”๋ก 
  • ์œ ํšจ ๋ฐ”์ธ๋” ์ด์ง„ ์„ ๋ณ„ ($K_d \le 100\text{nM}$): ํ™œ์„ฑ ์•ฝ๋ฌผ ๊ธฐ์ค€์น˜($pK_d \ge 7.0$)๋ฅผ ์ถฉ์กฑํ•˜๋Š” ์œ ํšจ ๊ฒฐํ•ฉ ์„ ๋„๋ฌผ์งˆ(Lead Compound)์„ ๋†’์€ ์‹ ๋ขฐ๋„(AUC 0.90+)๋กœ ๋ถ„๋ฅ˜
  • ์„œ์—ด ์ˆœ์œ„ ๋ณด์กด ๋žญํ‚น (Concordance Index): ํ™”ํ•ฉ๋ฌผ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ ๋‚ด ์šฐ์„ ์ˆœ์œ„ ๋žญํ‚น์„ ์ •ํ™•ํ•˜๊ฒŒ ๋ณด์กด(CI 0.89+)ํ•˜์—ฌ ์‹คํ—˜์‹ค ํ•ฉ์„ฑ(Wet-lab) ๊ฒ€์ฆ ๋น„์šฉ ๋Œ€ํญ ์ ˆ๊ฐ
  • ๋ฌด์„œ๋ฒ„(Serverless) ๋ธŒ๋ผ์šฐ์ € ์Šคํฌ๋ฆฌ๋‹: ONNX ๋ฐ .baar3 Direct-mmap ์—”์ง„์„ ํ†ตํ•ด ๊ณ ๊ฐ€ GPU ์ธํ”„๋ผ ์—†์ด ์ผ๋ฐ˜ ๋žฉํ†ฑ ๋ธŒ๋ผ์šฐ์ €(WebGPU/WASM)์—์„œ ์ฆ‰์‹œ ๊ฐ€์ƒ ์Šคํฌ๋ฆฌ๋‹ ์ˆ˜ํ–‰

๐Ÿ“Š Live Kinome Benchmark & Throughput (์‹ค์ „ ๋ฒค์น˜๋งˆํฌ ๋ฐ ์†๋„ ๋น„๊ต)

ํ•˜๋ฒ„๋“œ TDC Davis Kinome ๋ฒค์น˜๋งˆํฌ ํ…Œ์ŠคํŠธ์…‹ ๋ฐ ๊ณต์ธ ๋ฒ ์ด์Šค๋ผ์ธ ๋ชจ๋ธ๊ณผ์˜ ์ •๋Ÿ‰์  ์‹ค์ธก ๋น„๊ต ๊ฒฐ๊ณผ์ž…๋‹ˆ๋‹ค.

Model Architecture Type Parameters Concordance Index (CI) โ†‘ Affinity RMSE โ†“ Screening Speed (Throughput) BAAR-Drug Speed Advantage
Bio-LLM (ESM-2 + Chem) Pretrained Bio-LLM ~150M+ 0.905 0.435 ~11 mol/s 300.0๋ฐฐ ๋น ๋ฆ„
GraphDTA (Nguyen et al.) Molecular GNN + CNN ~1.5M 0.892 0.478 ~80 mol/s 41.3๋ฐฐ ๋น ๋ฆ„
MolTrans (Huang et al.) Substructure Transformer ~5.0M 0.890 0.473 ~125 mol/s 26.4๋ฐฐ ๋น ๋ฆ„
DeepDTA (ร–ztรผrk et al.) Standard 1D-CNN ~0.8M 0.878 0.511 ~280 mol/s 11.8๋ฐฐ ๋น ๋ฆ„
BAAR3-drug-dti (Ours) BioInception + GAU + Cross-Attn ~0.90M 0.889 ~ 0.894 0.462 ~ 0.485 > 3,300 mol/s ๊ธฐ์ค€ ์—”์ง„ (1.0x)

โ€ป ์‹ค์ธก ์ฒ˜๋ฆฌ๋Ÿ‰(Throughput)์€ ๋‹จ์ผ ์›Œํฌ์Šคํ…Œ์ด์…˜ ํ™˜๊ฒฝ์—์„œ ๋™์ผ ๋ฐฐ์น˜ ํฌ๊ธฐ๋กœ ์ธก์ •ํ•œ ์ดˆ๋‹น ๋ถ„์ž-ํ‘œ์  ๊ฒฐํ•ฉ ํ‰๊ฐ€ ๊ฑด์ˆ˜์ž…๋‹ˆ๋‹ค.


๐Ÿ“ฆ Deployment Artifacts (๋ฐฐํฌ ์•„ํ‹ฐํŒฉํŠธ ๊ทœ๊ฒฉ)

  1. model_dti_fp16.baar3: Zero-Copy ๋ฉ”๋ชจ๋ฆฌ ๋งคํ•‘ ๊ธฐ๋ฐ˜ ๊ณ ์† ์ถ”๋ก  ์ปค๋„ (1.8 MiB)
  2. model_dti_q8.baar3: ๋ชจ๋ฐ”์ผ/์—ฃ์ง€ ์ดˆ์ €์šฉ๋Ÿ‰ 8๋น„ํŠธ ๋Œ€์นญ ์–‘์žํ™” ๋ฐ”์ด๋„ˆ๋ฆฌ (0.86 MiB)
  3. model_dti.onnx: ์›น ๋ธŒ๋ผ์šฐ์ €(WASM/WebGPU) ๋ฐ ๋ฐฑ์—”๋“œ ํ†ตํ•ฉ์šฉ ONNX ๋Ÿฐํƒ€์ž„ ๊ทธ๋ž˜ํ”„
  4. model.safetensors: PyTorch ํ‘œ์ค€ ๊ฐ€์ค‘์น˜ ํŒŒ์ผ
  5. vocab.json: 20๊ฐœ ์•„๋ฏธ๋…ธ์‚ฐ ๋ฐ ๋ถ„์ž SMILES ํ† ํฐ ์ธ๋ฑ์Šค ์‚ฌ์ „
  6. dti_test_evaluation_report.png: Parity Plot, Residuals, ROC-PR Curve, Throughput 4ํŒจ๋„ ์ข…ํ•ฉ ํ•™์ˆ  ๊ฒ€์ฆ ๋ฆฌํฌํŠธ

๐Ÿ“„ License & Commercial Terms (์—„๊ฒฉํ•œ ์ „๋ฉด ์œ ๋ฃŒ ๋ผ์ด์„ ์Šค ์ •์ฑ…)

โš ๏ธ ๋ณธ ๋ชจ๋ธ์€ ํ•™์ˆ  ์—ฐ๊ตฌ์™€ ์ƒ์—…์  ์ด์šฉ ๋ชจ๋‘ ์œ ๋ฃŒ ๋ผ์ด์„ ์Šค ๊ณ„์•ฝ ์ฒด๊ฒฐ์ด ํ•„์ˆ˜์ ์ธ ์ „๋ฉด ์œ ๋ฃŒ(Proprietary Paid) ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.

1. Mandatory Paid License (์—ฐ๊ตฌยทํ•™์ˆ  ๋ชฉ์ ์„ ํฌํ•จํ•œ ์ „๋ฉด ์œ ๋ฃŒ ๊ณ„์•ฝ ํ•„์ˆ˜)

  • No Free Usage (๋ฌด๋ฃŒ ์ด์šฉ ๋ถˆ๊ฐ€):
    This model is proprietary technology. Free use is strictly NOT permitted for any purpose, including non-commercial academic research, university studies, educational evaluations, or non-profit projects.
    ๋ณธ ๋ชจ๋ธ์€ ๋…์  ๊ธฐ์ˆ  ์ž์‚ฐ์œผ๋กœ, ๋น„์˜๋ฆฌ ํ•™์ˆ  ์—ฐ๊ตฌ, ๋Œ€ํ•™ ๋ฐ ๊ต์œก ๊ธฐ๊ด€์˜ ์‹œํ—˜ ํ‰๊ฐ€, ๋…ผ๋ฌธ ์ž‘์„ฑ, ์—ฐ๊ตฌ์†Œ ํ”„๋กœ์ ํŠธ๋ฅผ ํฌํ•จํ•œ ์ผ์ฒด์˜ ๋ฌด๋ฃŒ ์ด์šฉ์ด ์—„๊ฒฉํžˆ ๊ธˆ์ง€๋ฉ๋‹ˆ๋‹ค.
  • Academic Research License (์—ฐ๊ตฌ์šฉ ์œ ๋ฃŒ ๋ผ์ด์„ ์Šค):
    Academic institutions, laboratories, and researchers must purchase an official Academic Research License prior to downloading, reproducing, fine-tuning, or executing this model.
    ๋Œ€ํ•™ ์—ฐ๊ตฌ์†Œ, ํ•™์ˆ  ๋‹จ์ฒด ๋ฐ ๊ฐœ์ธ ์—ฐ๊ตฌ์ž๊ฐ€ ๋ณธ ๋ชจ๋ธ์„ ๋‹ค์šด๋กœ๋“œ, ์žฌํ˜„, ํŒŒ์ธํŠœ๋‹, ๊ฐ€์ƒ ์Šคํฌ๋ฆฌ๋‹ ์—ฐ๊ตฌ์— ์‚ฌ์šฉํ•˜๊ณ ์ž ํ•˜๋Š” ๊ฒฝ์šฐ์—๋„ ๋ฐ˜๋“œ์‹œ ์‚ฌ์ „์— ์—ฐ๊ตฌ์šฉ ์œ ๋ฃŒ ๋ผ์ด์„ ์Šค๋ฅผ ๋ฐœ๊ธ‰๋ฐ›์œผ์…”์•ผ ํ•ฉ๋‹ˆ๋‹ค.

2. Commercial Enterprise Licensing (๊ธฐ์—… ์ƒ์šฉํ™” ๋ฐ ์‹ ์•ฝ ํŒŒ์ดํ”„๋ผ์ธ ๋„์ž…)

  • Biotech & Pharma Deployment (์ œ์•ฝยท๋ฐ”์ด์˜ค ๊ธฐ์—… ๋„์ž…):
    Any pharmaceutical company, biotech startup, CRO, or commercial entity integrating this model into virtual screening pipelines, drug candidate discovery, or commercial APIs must acquire an Enterprise Commercial License.
    ์‹ ์•ฝ ๊ฐœ๋ฐœ ํŒŒ์ดํ”„๋ผ์ธ, ๊ฐ€์ƒ ์Šคํฌ๋ฆฌ๋‹ ์†”๋ฃจ์…˜, ์œ ๋ฃŒ API ์„œ๋น„์Šค ๋ฐ ํ”Œ๋žซํผ์— ๋ณธ ์—”์ง„์„ ํƒ‘์žฌํ•˜๊ณ ์ž ํ•˜๋Š” ๊ธฐ์—…์€ ๊ธฐ์—…์šฉ ์ƒ์šฉ ๋ผ์ด์„ ์Šค๋ฅผ ์ฒด๊ฒฐํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
    • ๋ผ์ด์„ ์Šค ๊ฒฌ์  ๋ฐ ๋ผ์ด์„ ์Šค ๋ฐœ๊ธ‰ ๋ฌธ์˜ (License Inquiries): admin@099.kr
    • ๊ณต์‹ ์›น์‚ฌ์ดํŠธ (Official Website): https://baar.uk/model/

โš ๏ธ Operational Notes (๊ฐ€์ƒ ์Šคํฌ๋ฆฌ๋‹ ํŒŒ์ดํ”„๋ผ์ธ ๊ฐ€์ด๋“œ)

  • Hierarchical Screening Cascade (๋‹จ๊ณ„์  ์Šคํฌ๋ฆฌ๋‹ ๊ถŒ์žฅ):
    ์ˆ˜๋ฐฑ๋งŒ ๊ฐœ ์ด์ƒ์˜ ๋Œ€๊ทœ๋ชจ ํ™”ํ•ฉ๋ฌผ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ(ZINC, Enamine ๋“ฑ) ํƒ์ƒ‰ ์‹œ, 1์ฐจ๋กœ ๋ณธ ์ดˆ๊ณ ์† ์—”์ง„(BAAR3-drug-dti)์„ ํ†ตํ•ด ์ƒ์œ„ 1%์˜ ์œ ํšจ ๊ฒฐํ•ฉ ํ›„๋ณด(pKd โ‰ฅ 7.0)๋ฅผ ์ดˆ๋‹จ์‹œ๊ฐ„์— ํ•„ํ„ฐ๋งํ•œ ํ›„, 2์ฐจ ์ •๋ฐ€ ๋ถ„์ž ๋„ํ‚น(AutoDock/Schrรถdinger)์œผ๋กœ ์—ฐ๊ณ„ํ•˜๋Š” ๊ณ„์ธต์  ํŒŒ์ดํ”„๋ผ์ธ ๊ตฌ์„ฑ์„ ์ ๊ทน ๊ถŒ์žฅํ•ฉ๋‹ˆ๋‹ค.
  • Sequence Length Truncation (์ž…๋ ฅ ๊ทœ๊ฒฉ):
    ๋‹จ๋ฐฑ์งˆ ์„œ์—ด ์ตœ๋Œ€ 600 ์•„๋ฏธ๋…ธ์‚ฐ, ์•ฝ๋ฌผ SMILES ์ตœ๋Œ€ 120์ž๋กœ ์‚ฌ์ „ ํ† ํฌ๋‚˜์ด์ง•๋˜์–ด ์—ฐ์‚ฐ ํšจ์œจ์„ ๊ทน๋Œ€ํ™”ํ•ฉ๋‹ˆ๋‹ค.
  • Client-Side Screening (๋ณด์•ˆ ๊ฐ•ํ™”):
    ๋™๋ด‰๋œ ONNX ๋ฐ .baar3 ํฌ๋งท์„ ํ™œ์šฉํ•˜๋ฉด, ๋น„๊ณต๊ฐœ ํ›„๋ณด ๋ฌผ์งˆ SMILES ๋ฐ์ดํ„ฐ๋ฅผ ์™ธ๋ถ€ ํด๋ผ์šฐ๋“œ๋กœ ์ „์†กํ•˜์ง€ ์•Š๊ณ  ์‚ฌ๋‚ด ๋กœ์ปฌ ๋ธŒ๋ผ์šฐ์ €๋‚˜ ๋…๋ฆฝ๋œ ๋ณด์•ˆ ์—ฃ์ง€ ํ™˜๊ฒฝ์—์„œ ์•ˆ์ „ํ•˜๊ฒŒ ์Šคํฌ๋ฆฌ๋‹ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

๐Ÿ’ผ Opportunities & Contact (์ฑ„์šฉ ์ œ์•ˆ ๋ฐ ๊ธฐ์ˆ  ํ˜‘๋ ฅ)

This project demonstrates practical competency in designing ultra-lightweight bioinformatics neural architectures, high-throughput in silico virtual screening pipelines, and hardware-optimized zero-copy deployment formats (.baar3, ONNX).

I am actively seeking AI Engineering / Computational Biology Research opportunities, senior recruitment offers, and technical partnerships.

  • ๊ณต์‹ ์›น์‚ฌ์ดํŠธ (Official Website): https://baar.uk/model/
  • ์—ฐ๊ตฌ ๋ฐ ๊ธฐ์—… ์œ ๋ฃŒ ๋ผ์ด์„ ์Šค ๋ฌธ์˜ (Paid Licensing): admin@099.kr
  • ์ฑ„์šฉ ๋ฐ ํฌ์ง€์…˜ ์ œ์•ˆ (Recruitment & Hiring): admin@099.kr
  • ์‹ ์•ฝ AI ๊ณต๋™ ์—ฐ๊ตฌ ๋ฐ ๊ธฐ์ˆ  ์ œํœด (Partnerships): admin@099.kr

์ˆ˜์–ต ์›๋Œ€์˜ ํด๋ผ์šฐ๋“œ GPU ์ธํ”„๋ผ ์—†์ด๋„ ์ผ๋ฐ˜ ์—ฃ์ง€ ๋ฐ ๋ธŒ๋ผ์šฐ์ €์—์„œ ์ดˆ๋‹น ์ˆ˜์ฒœ ๊ฐœ์˜ ์•ฝ๋ฌผ-๋‹จ๋ฐฑ์งˆ ๊ฒฐํ•ฉ์„ ์‹ค์‹œ๊ฐ„ ํŒ๋ณ„ํ•˜๋Š” ์ดˆ๊ฒฝ๋Ÿ‰ยท๊ณ ํšจ์œจ ๋ฐ”์ด์˜ค AI ์•„ํ‚คํ…์ฒ˜๋ฅผ ์ง์ ‘ ์„ค๊ณ„ํ•˜๊ณ  ์ตœ์ ํ™” ๋ฐฐํฌํ•˜๋Š” AI ์—”์ง€๋‹ˆ์–ด์ž…๋‹ˆ๋‹ค. ์ •๊ทœ์ง ํฌ์ง€์…˜ ์ œ์•ˆ ๋ฐ ๋ผ์ด์„ ์Šค ๋„์ž… ๋…ผ์˜๋ฅผ ํ™˜์˜ํ•ฉ๋‹ˆ๋‹ค.

Downloads last month
-
Safetensors
Model size
896k params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Collection including aixk/BAAR3-drug-dti