redesign: clean header banner + consistent sentence-case About text
Browse files
app.py
CHANGED
|
@@ -132,7 +132,7 @@ CUSTOM_CSS = """
|
|
| 132 |
.dark .tabs { background: #ffffff !important; }
|
| 133 |
.dark .tab-nav button { color: #2d3748 !important; }
|
| 134 |
.dark .tab-nav button.selected {
|
| 135 |
-
color: #
|
| 136 |
border-color: #3182ce !important;
|
| 137 |
}
|
| 138 |
.dark .block { background: #ffffff !important; }
|
|
@@ -184,35 +184,46 @@ def _base_layout(**overrides) -> dict:
|
|
| 184 |
|
| 185 |
|
| 186 |
def build_header(last_updated: str, n_entries: int) -> str:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 187 |
return f"""
|
| 188 |
-
<div style="background:
|
| 189 |
-
|
| 190 |
-
margin-bottom:0.
|
| 191 |
-
|
| 192 |
-
|
| 193 |
-
|
| 194 |
-
|
| 195 |
-
<
|
| 196 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 197 |
<a href="{PAPER_URL}" target="_blank"
|
| 198 |
-
style="background:
|
| 199 |
-
|
| 200 |
-
text-decoration:none;font-size:0.85rem;
|
| 201 |
-
font-weight:600">\U0001f4c4 Paper</a>
|
| 202 |
<a href="{GITHUB_URL}" target="_blank"
|
| 203 |
-
style="background:
|
| 204 |
-
|
| 205 |
-
text-decoration:none;font-size:0.85rem;
|
| 206 |
-
font-weight:600">\U0001f4bb GitHub</a>
|
| 207 |
<a href="{HF_URL}" target="_blank"
|
| 208 |
-
style="background:
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 212 |
</div>
|
| 213 |
-
<div style="font-size:0.8rem;opacity:0.6;margin-top:0.5rem">
|
| 214 |
-
Romero Lab, Duke University · Last updated: {last_updated}
|
| 215 |
-
· 76 tasks · {n_entries} conditions</div>
|
| 216 |
</div>"""
|
| 217 |
|
| 218 |
|
|
@@ -278,9 +289,9 @@ def build_leaderboard_table(
|
|
| 278 |
"font-size:0.9rem"
|
| 279 |
)
|
| 280 |
TH = (
|
| 281 |
-
"background:#
|
| 282 |
-
"text-align:left;font-size:0.
|
| 283 |
-
"letter-spacing:0.
|
| 284 |
)
|
| 285 |
|
| 286 |
rows = []
|
|
@@ -300,7 +311,7 @@ def build_leaderboard_table(
|
|
| 300 |
else:
|
| 301 |
llm_rank += 1
|
| 302 |
rcolor = {1: "#d69e2e", 2: "#a0aec0", 3: "#c17832"}.get(
|
| 303 |
-
llm_rank, "#
|
| 304 |
)
|
| 305 |
rsize = (
|
| 306 |
"1.1rem"
|
|
@@ -407,8 +418,8 @@ def build_heatmap(entry: dict) -> str:
|
|
| 407 |
"""HTML heatmap table for one agent across 17 taxonomy cells."""
|
| 408 |
ts = entry.get("taxonomy_scores", {})
|
| 409 |
TH = (
|
| 410 |
-
"background:#
|
| 411 |
-
"text-align:center;font-size:0.75rem"
|
| 412 |
)
|
| 413 |
TD = (
|
| 414 |
"text-align:center;padding:0.5rem;font-size:0.85rem;"
|
|
@@ -520,7 +531,7 @@ def build_mode_cards(entries: list) -> str:
|
|
| 520 |
cards.append(
|
| 521 |
'<div style="background:white;border-radius:10px;padding:1.2rem;'
|
| 522 |
'box-shadow:0 1px 3px rgba(0,0,0,0.08)">'
|
| 523 |
-
f'<h4 style="font-size:0.95rem;color:#
|
| 524 |
f'margin:0 0 0.8rem">{name}</h4>'
|
| 525 |
f'{"".join(lines)}</div>'
|
| 526 |
)
|
|
@@ -536,148 +547,152 @@ def build_mode_cards(entries: list) -> str:
|
|
| 536 |
|
| 537 |
|
| 538 |
def build_about() -> str:
|
| 539 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 540 |
<div style="max-width:900px;margin:0 auto">
|
| 541 |
|
| 542 |
-
<div
|
| 543 |
-
|
| 544 |
-
<
|
| 545 |
-
What is BioDesignBench?</h2>
|
| 546 |
-
<p style="margin-bottom:0.8rem;color:#2d3748;line-height:1.6">
|
| 547 |
BioDesignBench is the first comprehensive benchmark for evaluating
|
| 548 |
LLM agents on protein design tasks via MCP (Model Context Protocol)
|
| 549 |
tool use. Unlike existing benchmarks that focus on model-only
|
| 550 |
-
|
| 551 |
-
<strong>
|
| 552 |
-
|
| 553 |
<div style="display:grid;grid-template-columns:
|
| 554 |
-
repeat(auto-fit,minmax(140px,1fr));gap:
|
| 555 |
-
|
| 556 |
-
|
| 557 |
-
<div style="font-size:1.8rem;font-weight:
|
| 558 |
76</div>
|
| 559 |
-
<div style="font-size:0.
|
| 560 |
</div>
|
| 561 |
-
<div
|
| 562 |
-
|
| 563 |
-
<div style="font-size:1.8rem;font-weight:700;color:#3182ce">
|
| 564 |
17</div>
|
| 565 |
-
<div style="font-size:0.
|
| 566 |
</div>
|
| 567 |
-
<div
|
| 568 |
-
|
| 569 |
-
<div style="font-size:1.8rem;font-weight:700;color:#3182ce">
|
| 570 |
17</div>
|
| 571 |
-
<div style="font-size:0.
|
| 572 |
</div>
|
| 573 |
-
<div
|
| 574 |
-
|
| 575 |
-
<div style="font-size:1.8rem;font-weight:700;color:#3182ce">
|
| 576 |
100</div>
|
| 577 |
-
<div style="font-size:0.
|
| 578 |
</div>
|
| 579 |
</div>
|
| 580 |
</div>
|
| 581 |
|
| 582 |
-
<div
|
| 583 |
-
|
| 584 |
-
<
|
| 585 |
-
|
| 586 |
-
|
| 587 |
-
|
| 588 |
-
|
| 589 |
-
|
| 590 |
-
|
| 591 |
-
|
| 592 |
-
|
| 593 |
-
|
| 594 |
-
<
|
| 595 |
-
|
| 596 |
-
|
| 597 |
-
|
| 598 |
-
|
| 599 |
-
|
| 600 |
-
|
| 601 |
-
line-height:1.5">POST /evaluate
|
| 602 |
-
|
| 603 |
-
Input:
|
| 604 |
-
{
|
| 605 |
-
"task_id": "dnb_sig_001",
|
| 606 |
"task_description": "Design a de novo binder for...",
|
| 607 |
"available_tools": [...],
|
| 608 |
"max_steps": 50,
|
| 609 |
"timeout_sec": 300
|
| 610 |
-
}
|
| 611 |
|
| 612 |
-
|
| 613 |
-
{
|
| 614 |
"sequences": ["MKKL..."],
|
| 615 |
"run_log": [...],
|
| 616 |
"total_steps": 12,
|
| 617 |
"total_time_sec": 142.5
|
| 618 |
-
}</pre>
|
| 619 |
-
<h3
|
| 620 |
-
|
| 621 |
-
|
| 622 |
-
|
| 623 |
-
|
| 624 |
-
|
| 625 |
-
<p style="color:#2d3748">
|
| 626 |
3 example tasks are publicly available for development and
|
| 627 |
testing.</p>
|
| 628 |
|
| 629 |
-
<h3
|
| 630 |
-
|
| 631 |
-
|
| 632 |
-
|
| 633 |
-
|
| 634 |
-
|
| 635 |
-
|
| 636 |
-
<h3
|
| 637 |
-
|
| 638 |
-
|
| 639 |
-
<li>Maximum 2 submissions per month</li>
|
| 640 |
-
<li>
|
| 641 |
-
<li>3 example tasks are
|
| 642 |
</ul>
|
| 643 |
</div>
|
| 644 |
|
| 645 |
-
<div
|
| 646 |
-
|
| 647 |
-
<
|
| 648 |
-
|
| 649 |
-
|
| 650 |
-
|
| 651 |
-
|
| 652 |
-
|
| 653 |
-
|
| 654 |
-
|
| 655 |
-
|
| 656 |
-
|
| 657 |
-
|
| 658 |
-
<p
|
| 659 |
-
<strong>Feasibility (15 pts)</strong> —
|
| 660 |
-
length, composition, biophysical
|
| 661 |
-
<p
|
| 662 |
-
<strong>Novelty (5 pts)</strong> —
|
| 663 |
-
reference (lower = more novel =
|
| 664 |
-
<p
|
| 665 |
-
<strong>Diversity (10 pts)</strong> —
|
| 666 |
-
of generated designs</p>
|
| 667 |
</div>
|
| 668 |
|
| 669 |
-
<div
|
| 670 |
-
|
| 671 |
-
<
|
| 672 |
-
|
| 673 |
-
|
| 674 |
-
|
| 675 |
-
|
| 676 |
-
|
| 677 |
-
|
| 678 |
-
|
| 679 |
-
year={2026}
|
| 680 |
-
}</pre>
|
| 681 |
</div>
|
| 682 |
|
| 683 |
</div>"""
|
|
@@ -981,17 +996,18 @@ def create_app() -> gr.Blocks:
|
|
| 981 |
with gr.Tab("\U0001f4e4 Submit"):
|
| 982 |
gr.HTML("""
|
| 983 |
<div style="max-width:700px;margin:0 auto;padding:1rem">
|
| 984 |
-
<h2 style="color:#
|
| 985 |
-
|
| 986 |
-
|
|
|
|
| 987 |
Submit your protein design agent for benchmarking.
|
| 988 |
Your agent must be hosted as a POST endpoint that accepts
|
| 989 |
-
task
|
| 990 |
-
<strong>You bear all LLM and
|
| 991 |
-
we only run Boltz structure prediction on our
|
| 992 |
-
<div style="background:#
|
| 993 |
-
padding:0.8rem;border-radius:
|
| 994 |
-
font-size:0.85rem;color:#
|
| 995 |
<strong>Rate limit:</strong> 2 submissions per calendar
|
| 996 |
month per organization.</div>
|
| 997 |
</div>""")
|
|
@@ -1065,9 +1081,10 @@ def create_app() -> gr.Blocks:
|
|
| 1065 |
with gr.Tab("\U0001f6e0 Status"):
|
| 1066 |
gr.HTML("""
|
| 1067 |
<div style="max-width:800px;margin:0 auto;padding:1rem">
|
| 1068 |
-
<h2 style="color:#
|
| 1069 |
-
|
| 1070 |
-
|
|
|
|
| 1071 |
Check your submission status or manage the pipeline
|
| 1072 |
(admin only).</p>
|
| 1073 |
</div>""")
|
|
@@ -1100,7 +1117,7 @@ def create_app() -> gr.Blocks:
|
|
| 1100 |
if sub.get("overall_score") is not None:
|
| 1101 |
score_html = (
|
| 1102 |
f'<div style="font-size:1.2rem;'
|
| 1103 |
-
f'font-weight:700;color:#
|
| 1104 |
f'margin-top:0.5rem">'
|
| 1105 |
f'Score: {sub["overall_score"]:.1f}/100'
|
| 1106 |
f'</div>'
|
|
@@ -1132,7 +1149,7 @@ def create_app() -> gr.Blocks:
|
|
| 1132 |
admin_msg = gr.HTML()
|
| 1133 |
|
| 1134 |
with admin_panel:
|
| 1135 |
-
gr.HTML('<h3 style="color:#
|
| 1136 |
'Pending Submissions</h3>')
|
| 1137 |
pending_html = gr.HTML()
|
| 1138 |
refresh_btn = gr.Button("Refresh List")
|
|
@@ -1150,7 +1167,7 @@ def create_app() -> gr.Blocks:
|
|
| 1150 |
)
|
| 1151 |
approve_msg = gr.HTML()
|
| 1152 |
|
| 1153 |
-
gr.HTML('<h3 style="color:#
|
| 1154 |
'Pipeline Control</h3>')
|
| 1155 |
with gr.Row():
|
| 1156 |
dispatch_id = gr.Textbox(
|
|
|
|
| 132 |
.dark .tabs { background: #ffffff !important; }
|
| 133 |
.dark .tab-nav button { color: #2d3748 !important; }
|
| 134 |
.dark .tab-nav button.selected {
|
| 135 |
+
color: #0f172a !important;
|
| 136 |
border-color: #3182ce !important;
|
| 137 |
}
|
| 138 |
.dark .block { background: #ffffff !important; }
|
|
|
|
| 184 |
|
| 185 |
|
| 186 |
def build_header(last_updated: str, n_entries: int) -> str:
|
| 187 |
+
btn = (
|
| 188 |
+
"display:inline-block;padding:0.45rem 1.1rem;border-radius:8px;"
|
| 189 |
+
"text-decoration:none;font-size:0.82rem;font-weight:600;"
|
| 190 |
+
"transition:opacity 0.15s"
|
| 191 |
+
)
|
| 192 |
return f"""
|
| 193 |
+
<div style="background:#ffffff;border:1px solid #e2e8f0;
|
| 194 |
+
padding:2.2rem 2rem 1.8rem;text-align:center;
|
| 195 |
+
border-radius:16px;margin-bottom:0.8rem;
|
| 196 |
+
box-shadow:0 1px 4px rgba(0,0,0,0.04)">
|
| 197 |
+
<p style="margin:0 0 0.3rem;font-size:0.75rem;font-weight:700;
|
| 198 |
+
letter-spacing:0.12em;text-transform:uppercase;
|
| 199 |
+
color:#3182ce">Romero Lab · Duke University</p>
|
| 200 |
+
<h1 style="font-size:2rem;margin:0;font-weight:800;color:#0f172a;
|
| 201 |
+
letter-spacing:-0.02em">
|
| 202 |
+
\U0001f9ec BioDesignBench</h1>
|
| 203 |
+
<p style="color:#64748b;margin:0.4rem 0 0;font-size:1rem;
|
| 204 |
+
font-weight:400">
|
| 205 |
+
Evaluating LLM agents on protein design via MCP tools</p>
|
| 206 |
+
<div style="margin-top:1rem;display:flex;justify-content:center;
|
| 207 |
+
gap:0.6rem;flex-wrap:wrap">
|
| 208 |
<a href="{PAPER_URL}" target="_blank"
|
| 209 |
+
style="{btn};background:#0f172a;color:#ffffff">
|
| 210 |
+
\U0001f4c4 Paper</a>
|
|
|
|
|
|
|
| 211 |
<a href="{GITHUB_URL}" target="_blank"
|
| 212 |
+
style="{btn};background:#f1f5f9;color:#334155">
|
| 213 |
+
\U0001f4bb GitHub</a>
|
|
|
|
|
|
|
| 214 |
<a href="{HF_URL}" target="_blank"
|
| 215 |
+
style="{btn};background:#f1f5f9;color:#334155">
|
| 216 |
+
\U0001f917 HuggingFace</a>
|
| 217 |
+
</div>
|
| 218 |
+
<div style="margin-top:1rem;display:flex;justify-content:center;
|
| 219 |
+
gap:1.5rem;flex-wrap:wrap">
|
| 220 |
+
<span style="font-size:0.78rem;color:#94a3b8">
|
| 221 |
+
76 tasks</span>
|
| 222 |
+
<span style="font-size:0.78rem;color:#94a3b8">
|
| 223 |
+
{n_entries} agents</span>
|
| 224 |
+
<span style="font-size:0.78rem;color:#94a3b8">
|
| 225 |
+
Updated {last_updated}</span>
|
| 226 |
</div>
|
|
|
|
|
|
|
|
|
|
| 227 |
</div>"""
|
| 228 |
|
| 229 |
|
|
|
|
| 289 |
"font-size:0.9rem"
|
| 290 |
)
|
| 291 |
TH = (
|
| 292 |
+
"background:#0f172a;color:white;padding:0.75rem 1rem;"
|
| 293 |
+
"text-align:left;font-size:0.75rem;text-transform:uppercase;"
|
| 294 |
+
"letter-spacing:0.05em;font-weight:600"
|
| 295 |
)
|
| 296 |
|
| 297 |
rows = []
|
|
|
|
| 311 |
else:
|
| 312 |
llm_rank += 1
|
| 313 |
rcolor = {1: "#d69e2e", 2: "#a0aec0", 3: "#c17832"}.get(
|
| 314 |
+
llm_rank, "#0f172a"
|
| 315 |
)
|
| 316 |
rsize = (
|
| 317 |
"1.1rem"
|
|
|
|
| 418 |
"""HTML heatmap table for one agent across 17 taxonomy cells."""
|
| 419 |
ts = entry.get("taxonomy_scores", {})
|
| 420 |
TH = (
|
| 421 |
+
"background:#0f172a;color:white;padding:0.6rem 0.8rem;"
|
| 422 |
+
"text-align:center;font-size:0.75rem;font-weight:600"
|
| 423 |
)
|
| 424 |
TD = (
|
| 425 |
"text-align:center;padding:0.5rem;font-size:0.85rem;"
|
|
|
|
| 531 |
cards.append(
|
| 532 |
'<div style="background:white;border-radius:10px;padding:1.2rem;'
|
| 533 |
'box-shadow:0 1px 3px rgba(0,0,0,0.08)">'
|
| 534 |
+
f'<h4 style="font-size:0.95rem;color:#0f172a;'
|
| 535 |
f'margin:0 0 0.8rem">{name}</h4>'
|
| 536 |
f'{"".join(lines)}</div>'
|
| 537 |
)
|
|
|
|
| 547 |
|
| 548 |
|
| 549 |
def build_about() -> str:
|
| 550 |
+
h2 = (
|
| 551 |
+
'style="color:#0f172a;margin:0 0 0.8rem;font-size:1.25rem;'
|
| 552 |
+
'font-weight:700"'
|
| 553 |
+
)
|
| 554 |
+
h3 = (
|
| 555 |
+
'style="color:#334155;margin:1.2rem 0 0.5rem;font-size:1rem;'
|
| 556 |
+
'font-weight:600"'
|
| 557 |
+
)
|
| 558 |
+
p = 'style="margin-bottom:0.8rem;color:#475569;line-height:1.6"'
|
| 559 |
+
card = (
|
| 560 |
+
'style="background:#ffffff;border:1px solid #e2e8f0;'
|
| 561 |
+
'border-radius:12px;padding:2rem;margin-bottom:1.2rem"'
|
| 562 |
+
)
|
| 563 |
+
stat_box = (
|
| 564 |
+
'style="background:#f8fafc;border:1px solid #e2e8f0;'
|
| 565 |
+
'border-radius:10px;padding:1rem;text-align:center"'
|
| 566 |
+
)
|
| 567 |
+
return f"""
|
| 568 |
<div style="max-width:900px;margin:0 auto">
|
| 569 |
|
| 570 |
+
<div {card}>
|
| 571 |
+
<h2 {h2}>What is BioDesignBench?</h2>
|
| 572 |
+
<p {p}>
|
|
|
|
|
|
|
| 573 |
BioDesignBench is the first comprehensive benchmark for evaluating
|
| 574 |
LLM agents on protein design tasks via MCP (Model Context Protocol)
|
| 575 |
tool use. Unlike existing benchmarks that focus on model-only
|
| 576 |
+
metrics, BioDesignBench tests the full agentic design loop:
|
| 577 |
+
<strong>prompt → design → validate →
|
| 578 |
+
iterate</strong>.</p>
|
| 579 |
<div style="display:grid;grid-template-columns:
|
| 580 |
+
repeat(auto-fit,minmax(140px,1fr));gap:0.8rem;
|
| 581 |
+
margin:1rem 0">
|
| 582 |
+
<div {stat_box}>
|
| 583 |
+
<div style="font-size:1.8rem;font-weight:800;color:#0f172a">
|
| 584 |
76</div>
|
| 585 |
+
<div style="font-size:0.78rem;color:#64748b">design tasks</div>
|
| 586 |
</div>
|
| 587 |
+
<div {stat_box}>
|
| 588 |
+
<div style="font-size:1.8rem;font-weight:800;color:#0f172a">
|
|
|
|
| 589 |
17</div>
|
| 590 |
+
<div style="font-size:0.78rem;color:#64748b">taxonomy cells</div>
|
| 591 |
</div>
|
| 592 |
+
<div {stat_box}>
|
| 593 |
+
<div style="font-size:1.8rem;font-weight:800;color:#0f172a">
|
|
|
|
| 594 |
17</div>
|
| 595 |
+
<div style="font-size:0.78rem;color:#64748b">MCP tools</div>
|
| 596 |
</div>
|
| 597 |
+
<div {stat_box}>
|
| 598 |
+
<div style="font-size:1.8rem;font-weight:800;color:#0f172a">
|
|
|
|
| 599 |
100</div>
|
| 600 |
+
<div style="font-size:0.78rem;color:#64748b">point rubric</div>
|
| 601 |
</div>
|
| 602 |
</div>
|
| 603 |
</div>
|
| 604 |
|
| 605 |
+
<div {card}>
|
| 606 |
+
<h2 {h2}>How to submit</h2>
|
| 607 |
+
<h3 {h3}>1. Build your agent</h3>
|
| 608 |
+
<p {p}>
|
| 609 |
+
Create a protein design agent that accepts tasks via our API.
|
| 610 |
+
You can use our 17 reference MCP tools as-is, extend them, or
|
| 611 |
+
build entirely custom tools.</p>
|
| 612 |
+
<h3 {h3}>2. Host an API endpoint</h3>
|
| 613 |
+
<p {p}>
|
| 614 |
+
Your agent must be accessible as a POST endpoint that accepts
|
| 615 |
+
task payloads and returns designed protein sequences.</p>
|
| 616 |
+
<h3 {h3}>API specification</h3>
|
| 617 |
+
<pre style="background:#0f172a;color:#e2e8f0;padding:1.2rem;
|
| 618 |
+
border-radius:10px;font-size:0.8rem;overflow-x:auto;
|
| 619 |
+
line-height:1.6">POST /api/run
|
| 620 |
+
|
| 621 |
+
Request:
|
| 622 |
+
{{
|
| 623 |
+
"task_id": "dnb_ab_001",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 624 |
"task_description": "Design a de novo binder for...",
|
| 625 |
"available_tools": [...],
|
| 626 |
"max_steps": 50,
|
| 627 |
"timeout_sec": 300
|
| 628 |
+
}}
|
| 629 |
|
| 630 |
+
Response:
|
| 631 |
+
{{
|
| 632 |
"sequences": ["MKKL..."],
|
| 633 |
"run_log": [...],
|
| 634 |
"total_steps": 12,
|
| 635 |
"total_time_sec": 142.5
|
| 636 |
+
}}</pre>
|
| 637 |
+
<h3 {h3}>3. Submit and wait</h3>
|
| 638 |
+
<p {p}>
|
| 639 |
+
We dispatch 73 hidden tasks to your endpoint and verify results
|
| 640 |
+
with Boltz structure prediction.
|
| 641 |
+
Maximum <strong>2 submissions per month</strong> per organization.</p>
|
| 642 |
+
<p {p}>
|
|
|
|
| 643 |
3 example tasks are publicly available for development and
|
| 644 |
testing.</p>
|
| 645 |
|
| 646 |
+
<h3 {h3}>Reference MCP tools</h3>
|
| 647 |
+
<p {p}>
|
| 648 |
+
We provide 17 reference MCP tools for protein design. You can use
|
| 649 |
+
them as-is, extend them, or build entirely custom tools.
|
| 650 |
+
<a href="#" style="color:#2563eb;font-weight:500">
|
| 651 |
+
GitHub repository →</a></p>
|
| 652 |
+
|
| 653 |
+
<h3 {h3}>Limits</h3>
|
| 654 |
+
<ul style="color:#475569;padding-left:1.5rem;margin-bottom:0.8rem;
|
| 655 |
+
line-height:1.7">
|
| 656 |
+
<li>Maximum 2 submissions per calendar month per organization</li>
|
| 657 |
+
<li>73 hidden tasks are used for ranking</li>
|
| 658 |
+
<li>3 public example tasks are available for development</li>
|
| 659 |
</ul>
|
| 660 |
</div>
|
| 661 |
|
| 662 |
+
<div {card}>
|
| 663 |
+
<h2 {h2}>Scoring rubric (100 points)</h2>
|
| 664 |
+
<p {p}>
|
| 665 |
+
<strong>Approach (20 pts)</strong> — design methodology
|
| 666 |
+
coverage across 10 functional categories
|
| 667 |
+
(backbone generation, sequence design, structure prediction, etc.)</p>
|
| 668 |
+
<p {p}>
|
| 669 |
+
<strong>Orchestration (15 pts)</strong> — pipeline ordering,
|
| 670 |
+
intermediate validation, and iteration quality</p>
|
| 671 |
+
<p {p}>
|
| 672 |
+
<strong>Quality (35 pts)</strong> — three-tier graduated
|
| 673 |
+
scoring based on structure confidence (pLDDT, pTM), interface
|
| 674 |
+
metrics (ipTM, i_pAE), and interface physics</p>
|
| 675 |
+
<p {p}>
|
| 676 |
+
<strong>Feasibility (15 pts)</strong> — valid amino acids,
|
| 677 |
+
length constraints, composition, and biophysical plausibility</p>
|
| 678 |
+
<p {p}>
|
| 679 |
+
<strong>Novelty (5 pts)</strong> — sequence identity to
|
| 680 |
+
reference (lower identity = more novel = higher score)</p>
|
| 681 |
+
<p {p}>
|
| 682 |
+
<strong>Diversity (10 pts)</strong> — number and pairwise
|
| 683 |
+
diversity of generated designs</p>
|
| 684 |
</div>
|
| 685 |
|
| 686 |
+
<div {card}>
|
| 687 |
+
<h2 {h2}>Citation</h2>
|
| 688 |
+
<pre style="background:#0f172a;color:#e2e8f0;padding:1.2rem;
|
| 689 |
+
border-radius:10px;font-size:0.8rem;
|
| 690 |
+
line-height:1.6">@article{{biodesignbench2026,
|
| 691 |
+
title={{BioDesignBench: Evaluating LLM Agents on
|
| 692 |
+
Protein Design via MCP Tools}},
|
| 693 |
+
author={{Kim, Jason et al.}},
|
| 694 |
+
year={{2026}}
|
| 695 |
+
}}</pre>
|
|
|
|
|
|
|
| 696 |
</div>
|
| 697 |
|
| 698 |
</div>"""
|
|
|
|
| 996 |
with gr.Tab("\U0001f4e4 Submit"):
|
| 997 |
gr.HTML("""
|
| 998 |
<div style="max-width:700px;margin:0 auto;padding:1rem">
|
| 999 |
+
<h2 style="color:#0f172a;margin:0 0 0.5rem;
|
| 1000 |
+
font-weight:700;font-size:1.25rem">
|
| 1001 |
+
Submit your agent</h2>
|
| 1002 |
+
<p style="color:#475569;margin-bottom:1rem;line-height:1.6">
|
| 1003 |
Submit your protein design agent for benchmarking.
|
| 1004 |
Your agent must be hosted as a POST endpoint that accepts
|
| 1005 |
+
task payloads and returns designed sequences.
|
| 1006 |
+
<strong>You bear all LLM and tool costs</strong> —
|
| 1007 |
+
we only run Boltz structure prediction on our side.</p>
|
| 1008 |
+
<div style="background:#fefce8;border-left:3px solid #ca8a04;
|
| 1009 |
+
padding:0.8rem 1rem;border-radius:6px;
|
| 1010 |
+
margin-bottom:1rem;font-size:0.85rem;color:#713f12">
|
| 1011 |
<strong>Rate limit:</strong> 2 submissions per calendar
|
| 1012 |
month per organization.</div>
|
| 1013 |
</div>""")
|
|
|
|
| 1081 |
with gr.Tab("\U0001f6e0 Status"):
|
| 1082 |
gr.HTML("""
|
| 1083 |
<div style="max-width:800px;margin:0 auto;padding:1rem">
|
| 1084 |
+
<h2 style="color:#0f172a;margin:0 0 0.5rem;
|
| 1085 |
+
font-weight:700;font-size:1.25rem">
|
| 1086 |
+
Submission status</h2>
|
| 1087 |
+
<p style="color:#475569;margin-bottom:0.5rem;line-height:1.6">
|
| 1088 |
Check your submission status or manage the pipeline
|
| 1089 |
(admin only).</p>
|
| 1090 |
</div>""")
|
|
|
|
| 1117 |
if sub.get("overall_score") is not None:
|
| 1118 |
score_html = (
|
| 1119 |
f'<div style="font-size:1.2rem;'
|
| 1120 |
+
f'font-weight:700;color:#0f172a;'
|
| 1121 |
f'margin-top:0.5rem">'
|
| 1122 |
f'Score: {sub["overall_score"]:.1f}/100'
|
| 1123 |
f'</div>'
|
|
|
|
| 1149 |
admin_msg = gr.HTML()
|
| 1150 |
|
| 1151 |
with admin_panel:
|
| 1152 |
+
gr.HTML('<h3 style="color:#0f172a">'
|
| 1153 |
'Pending Submissions</h3>')
|
| 1154 |
pending_html = gr.HTML()
|
| 1155 |
refresh_btn = gr.Button("Refresh List")
|
|
|
|
| 1167 |
)
|
| 1168 |
approve_msg = gr.HTML()
|
| 1169 |
|
| 1170 |
+
gr.HTML('<h3 style="color:#0f172a;margin-top:1rem">'
|
| 1171 |
'Pipeline Control</h3>')
|
| 1172 |
with gr.Row():
|
| 1173 |
dispatch_id = gr.Textbox(
|