๐Ÿ”ฌ ADMET-AI๋ž€?

ADMET-AI ์•ฝ๋™ํ•™ ๋…์„ฑ ์˜ˆ์ธก ์‹ ์•ฝ๊ฐœ๋ฐœ SMILES ํ™”ํ•ฉ๋ฌผ ๋ถ„์„ Multi-task learning

ADMET-AI๋Š” ์‹ ์•ฝ๊ฐœ๋ฐœ ์ดˆ๊ธฐ ๋‹จ๊ณ„์—์„œ ํ™”ํ•ฉ๋ฌผ์˜ ADMET(ํก์ˆ˜, ๋ถ„ํฌ, ๋Œ€์‚ฌ, ๋ฐฐ์„ค, ๋…์„ฑ) ํŠน์„ฑ์„ ๋น ๋ฅด๊ฒŒ ์˜ˆ์ธกํ•ด์ฃผ๋Š” ์ธ๊ณต์ง€๋Šฅ ๊ธฐ๋ฐ˜ ๋ถ„์„ ๋„๊ตฌ์ž…๋‹ˆ๋‹ค. Multi-task ํ•™์Šต ๊ธฐ๋ฐ˜์˜ ๊ณ ์† ์˜ˆ์ธก์œผ๋กœ ๋…์„ฑ ๋ฐ ๋Œ€์‚ฌ ์•ˆ์ •์„ฑ ํ‰๊ฐ€์— ๊ฐ•์ ์„ ๋ณด์ž…๋‹ˆ๋‹ค.

0%
๋ชฉ์ฐจ
๋ชฉ์ฐจ

ADMET-AI ๋Š” ์‹ ์•ฝ๊ฐœ๋ฐœ ๊ณผ์ •์—์„œ ์ค‘์š”ํ•œ ADMET(Absorption, Distribution, Metabolism, Excretion, Toxicity) ํŠน์„ฑ์„ ์˜ˆ์ธกํ•˜๋Š” AI ๊ธฐ๋ฐ˜ ๋ถ„์„ ํ”Œ๋žซํผ์ž…๋‹ˆ๋‹ค. ์•ฝ๋ฌผ ํ›„๋ณด๋ฌผ์งˆ์˜ ์•ˆ์ „์„ฑ๊ณผ ์•ฝ๋ฆฌํ•™์  ์„ฑ์งˆ์„ ์‚ฌ์ „ ํ‰๊ฐ€ ํ•˜์—ฌ, ๊ฐœ๋ฐœ ์ดˆ๊ธฐ ๋‹จ๊ณ„์—์„œ ์œ ๋งํ•œ ํ™”ํ•ฉ๋ฌผ์„ ์„ ๋ณ„ํ•˜๋Š” ๋ฐ ๋„์›€์„ ์ค๋‹ˆ๋‹ค.

์ฐธ๊ณ : ADMET-AI๋Š” ํ™”ํ•ฉ๋ฌผ์˜ ๊ทธ๋ž˜ํ”„ ๊ตฌ์กฐ์™€ molecular fingerprint๋ฅผ ํ†ตํ•ฉํ•˜์—ฌ Multi-task learning ๊ธฐ๋ฐ˜์œผ๋กœ ์—ฌ๋Ÿฌ ํŠน์„ฑ์„ ๋™์‹œ์— ์˜ˆ์ธกํ•ฉ๋‹ˆ๋‹ค.

๐Ÿงช ์šฐ๋ฆฌ ํ”Œ๋žซํผ์—์„œ ์–ด๋–ป๊ฒŒ ์ œ๊ณต๋˜๋‚˜์š”?

์šฐ๋ฆฌ ํ”Œ๋žซํผ์—์„œ๋Š” ADMET-AI๋ฅผ ๋น ๋ฅด๊ณ  ์ง๊ด€์ ์ธ ๋ฐฉ์‹์œผ๋กœ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. SMILES ํ˜•์‹์˜ ํ™”ํ•ฉ๋ฌผ ์ •๋ณด๋ฅผ ์ž…๋ ฅํ•˜๋ฉด, ์ฃผ์š” ADMET property ์˜ˆ์ธก ๊ฒฐ๊ณผ์™€ ์‹œ๊ฐํ™” ์ž๋ฃŒ(heatmap, plot) ๋ฅผ ํ•จ๊ป˜ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

๋ณ„๋„์˜ ํŒŒ๋ผ๋ฏธํ„ฐ ์„ค์ • ์—†์ด๋„ ์‹ ์†ํ•œ ๋ถ„์„ ๊ฒฐ๊ณผ ๋ฅผ ์ œ๊ณตํ•˜๋ฏ€๋กœ, hit-to-lead ๊ณผ์ •์—์„œ ๋น ๋ฅธ ์˜์‚ฌ๊ฒฐ์ •์ด ๊ฐ€๋Šฅํ•ฉ๋‹ˆ๋‹ค.

๐Ÿ“ ์‚ฌ์šฉ ๋ฐฉ๋ฒ• ์•ˆ๋‚ด

1. ์ž‘์—… ์ด๋ฆ„ ์ง€์ •

  • ์˜ˆ: ADMETAI_test1

2. ์ž…๋ ฅ ๋ฐฉ์‹ ์„ ํƒ

  • ํ™”ํ•ฉ๋ฌผ ๊ตฌ์กฐ ์ž…๋ ฅ (SMILES ํฌ๋งท)

3. ํŒŒ๋ผ๋ฏธํ„ฐ ์„ค์ •

  • ๋ณ„๋„ ์„ค์ • ์—†์Œ (๊ธฐ๋ณธ๊ฐ’ ์ž๋™ ์ ์šฉ)

4. ๊ฒฐ๊ณผ ํ™•์ธ

  • ์˜ˆ์ธก๋œ ADMET ํŠน์„ฑ ๊ฐ’ (์˜ˆ: ํก์ˆ˜์œจ, ๋Œ€์‚ฌ ์•ˆ์ •์„ฑ, ๋…์„ฑ ๋“ฑ)
  • Confidence score ๋ฐ ์˜ˆ์ธก๊ฐ’ ranking
  • ์‹œ๊ฐํ™” ์ž๋ฃŒ (heatmap, plot ๋“ฑ)
Job Name: ADMETAI_test1
์ž…๋ ฅ๊ฐ’ ์˜ˆ์‹œ: CC(=O)OC1=CC=CC=C1C(=O)O

๐Ÿงฌ ๋ถ„์„ ๊ฒฐ๊ณผ ํ™œ์šฉ ์˜ˆ์‹œ

  • ์‹ ์•ฝ ํ›„๋ณด๋ฌผ์งˆ ํ•„ํ„ฐ๋ง: ํก์ˆ˜์œจ์ด๋‚˜ ๋…์„ฑ์ด ๊ธฐ์ค€ ์ดํ•˜์ธ ๋ฌผ์งˆ ์ œ์™ธ
  • Virtual screening ํ›„์† ๋ถ„์„: Diffdock ๋“ฑ ๊ฒฐํ•ฉ ์˜ˆ์ธก ๋„๊ตฌ์™€ ๊ฒฐํ•ฉํ•˜์—ฌ ์œ ๋ง ๋ฌผ์งˆ ์„ ๋ณ„
  • ์•ฝ๋ฌผ๋™ํƒœ ์˜ˆ์ธก: ๋Œ€์‚ฌ ๋ฐ˜์‘์„ฑ ๋˜๋Š” ์ฒด๋‚ด ๋ถ„ํฌ ๊ฒฝํ–ฅ ๋“ฑ ์˜ˆ์ธก

โš ๏ธ ์ฃผ์˜์‚ฌํ•ญ

ํ•ญ๋ชฉ๋‚ด์šฉ
์ž…๋ ฅ ํ˜•์‹SMILES ํ˜•์‹ ํ•„์ˆ˜
ํ•„์ˆ˜ ์ž…๋ ฅํ•˜๋‚˜ ์ด์ƒ์˜ ์œ ํšจํ•œ ํ™”ํ•ฉ๋ฌผ ๊ตฌ์กฐ
๋ถ„์„ ์‹œ๊ฐ„๋Œ€๋ถ€๋ถ„ ์ˆ˜ ์ดˆ~์ˆ˜ ๋ถ„ ๋‚ด ์ฒ˜๋ฆฌ ๊ฐ€๋Šฅ
ํ•ด์„ ์œ ์˜์ Confidence score๊ฐ€ ๋†’๋”๋ผ๋„ ์‹คํ—˜์  validation ํ•„์š”

โœ… ๋งˆ๋ฌด๋ฆฌ

ADMET-AI๋Š” ๋น ๋ฅด๊ณ  ์‹ ๋ขฐ๋„ ๋†’์€ ADMET ์˜ˆ์ธก์„ ์ง€์›ํ•˜๋Š” ์ตœ์‹  AI ๊ธฐ๋ฐ˜ ๋„๊ตฌ ์ž…๋‹ˆ๋‹ค. DILI, Amesformer, hERG-prediction ๋“ฑ์˜ ๋…์„ฑ ์˜ˆ์ธก ๋„๊ตฌ์™€ ๊ต์ฐจ ๊ฒ€์ฆ ํ•˜๊ฑฐ๋‚˜, Diffdock๊ณผ ์—ฐ๊ณ„ํ•˜์—ฌ ๊ฒฐํ•ฉ๋ ฅ๊ณผ ADMET ํŠน์„ฑ์„ ๋™์‹œ์— ๊ณ ๋ ค ํ•œ ์˜์‚ฌ๊ฒฐ์ •์—๋„ ํ™œ์šฉ ๊ฐ€๋Šฅํ•ฉ๋‹ˆ๋‹ค.

์ง€๊ธˆ ๋ฐ”๋กœ ADMET-AI ๋ถ„์„ ํŽ˜์ด์ง€์—์„œ ์ง์ ‘ ์‚ฌ์šฉํ•ด๋ณด์„ธ์š”!


ยฉ 2025. All rights reserved.