-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
2106 lines (2070 loc) · 115 KB
/
Copy pathindex.html
File metadata and controls
2106 lines (2070 loc) · 115 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<meta name="description" content="PDFBench">
<meta name="keywords" content="PDFBench">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>PDFBench: A Benchmark for De novo Protein Design from Function</title>
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-PYVRSFMDRL"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() {
dataLayer.push(arguments);
}
gtag('js', new Date());
gtag('config', 'G-PYVRSFMDRL');
</script>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="stylesheet" href="./static/css/bulma.min.css">
<link rel="stylesheet" href="./static/css/bulma-carousel.min.css">
<link rel="stylesheet" href="./static/css/bulma-slider.min.css">
<link rel="stylesheet" href="./static/css/fontawesome.all.min.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="./static/css/index.css">
<link rel="stylesheet" href="./static/css/leaderboard.css">
<link rel="icon" sizes="144x144" href="./static/images/PDFBench1.2.png">
<link rel="apple-touch-icon" sizes="144x144" href="./static/images/PDFBench1.2.png">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script defer src="./static/js/fontawesome.all.min.js"></script>
<script src="./static/js/bulma-carousel.min.js"></script>
<script src="./static/js/bulma-slider.min.js"></script>
<script src="./static/js/index.js"></script>
</head>
<body>
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-1 publication-title small-caps">
<img src="./static/images/PDFBench1.2.png" height="8%" width="8%" alt="Logo">
<span class="pdfbench">PDFBench</span>
</h1>
<h3 class="subtitle is-3 publication-subtitle">A Benchmark for <em>De novo</em> Protein Design
from Function</h3>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://github.com/Knlife">Jiahao Kuang</a><sup>1</sup>,</span>
<span class="author-block">
<a href="https://github.com/sornkL">Nuowei Liu</a><sup>1</sup>,</span>
<span class="author-block">
<a href="https://github.com/sornkL">Jie Wang</a><sup>1</sup>,</span>
<span class="author-block">
<a href="https://pdfbench.github.io">Changzhi Sun</a><sup>2</sup>,
</span>
<span class="author-block">
<a href="https://pdfbench.github.io">Tao Ji</a><sup>3</sup>,
</span>
<span class="author-block">
<a href="https://ybwu.org">Yuanbin Wu</a><sup>1</sup>,
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>School of Computer Science and Technology, East China
Normal University</span><br />
<span class="author-block"><sup>2</sup>Institute of Artificial Intelligence (TeleAI), China
Telecom</span><br />
<span class="author-block"><sup>3</sup>School of Computer Science, Fudan University</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- PDF Link. -->
<span class="link-block">
<a href="https://arxiv.org/abs/2505.20346"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Paper</span>
</a>
</span>
<!-- Code Link. -->
<span class="link-block">
<a href="https://github.com/pdfbench/PDFBench"
class="external-link button is-normal is-rounded is-dark">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
</span>
<!-- Dataset Link. -->
<span class="link-block">
<a href=" https://huggingface.co/datasets/Knlife/SwiwwProtIPG"
class="external-link button is-normal is-rounded is-dark">
<!-- <span class="icon">
<i class="far fa-images"></i>
</span> -->
<span>🤗 Dataset</span>
</a>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Abstract -->
<section class="section" id="abstract">
<div class="container is-max-desktop">
<!-- Abstract. -->
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-3">Abstract</h2>
<div class="content has-text-justified">
Function-guided protein design is a crucial task with significant applications in drug discovery and enzyme engineering. However, the field lacks a unified and comprehensive evaluation framework. Current models are assessed using inconsistent and limited subsets of metrics, which prevents fair comparison and a clear understanding of the relationships between different evaluation criteria. To address this gap, we introduce <img
src="./static/images/PDFBench1.2.png" height="3%" width="3%" alt="Logo" style="vertical-align: middle"><span class="pdfbench">PDFBench</span>, the first comprehensive benchmark for function-guided <em>de novo</em> protein design. Our benchmark systematically evaluates eight state-of-the-art models on 16 metrics across two key settings: description-guided design, for which we repurpose the Mol-Instructions dataset, originally lacking quantitative benchmarking, and keyword-guided design, for which we introduce a new test set, <em>SwissTest</em>, created with a strict datetime cutoff to ensure data integrity. By benchmarking across a wide array of metrics and analyzing their correlations, <img
src="./static/images/PDFBench1.2.png" height="3%" width="3%" alt="Logo" style="vertical-align: middle"><span class="pdfbench">PDFBench</span> enables more reliable model comparisons and provides key insights to guide future research.
</div>
</div>
</div>
<img id="fig-task-overview" src="static/images/introduction/task_overview.png" alt="Task Overview">
<div style="text-align: center;font-size: 1rem;margin-bottom: 2%;font-weight:550;">
<strong>Figure 1:</strong>
Examples of inputs and outputs for the description-guided protein design task and the keyword-guided protein design task (using GO and/or IPR keywords as inputs). Note that the GO and IPR terms can be converted into textual descriptions.
</div>
<img id="fig-metric-coverage" src="static/images/introduction/merge.png" alt="Metric Coverage">
<div style="text-align: center;font-size: 1rem;margin-bottom: 2%;font-weight:550;">
<strong>Figure 2:</strong>
Overview of current function-guided protein design models evaluated using different metrics, highlighting the lack of a unified and comprehensive evaluation framework. (a) Proportion of metrics employed in each previous work. In <img src="./static/images/PDFBench1.2.png" height="3%" width="3%" alt="Logo" style="vertical-align: middle"><span class="pdfbench">PDFBench</span>, metrics are categorized into 6 dimensions, and we show that none of the prior works have been evaluated across all dimensions. (b) Detailed view of the metrics in <img src="./static/images/PDFBench1.2.png" height="3%" width="3%" alt="Logo" style="vertical-align: middle"><span class="pdfbench">PDFBench</span>, with several representative metrics from each dimension presented
</div>
<!--/ Abstract. -->
</div>
</section>
<!-- Description-guided -->
<section class="section" id="results-description">
<div class="container is-max-desktop">
<div class="is-centered has-text-centered">
<h1 class="title is-3">Description-guided</h1>
<div class="content has-text-justified">
<p>
In <a href="#res-desc-seq">Table 5</a>, <a href="#results-desc-stc">Table 6</a> and <a href="#res-desc-alt">Table 7</a>, we report the benchmark results for the description-guided task on all 16 metrics. Findings are summarized as follows:
</p>
<ul>
<li><strong>ProDVa can design relatively plausible sequence.</strong> Good sequence plausibility is fundamental to foldability and language alignment. The sequences designed by ProDVa exhibit repeat scores exceeding those of natural proteins and suboptimal perplexity scores in sequence rationality, indicating that ProDVa's design modules are capable of generating reasonable protein sequences.</li>
<li>
<strong>
ProDVa and Pinal Generate Foldable Proteins.
</strong>
Both ProDVa and Pinal achieve substantially higher foldability scores compared with all other models. Specifically, ProDVa reaches the best pLDDT (76.86) and lowest PAE (8.66), while Pinal follows closely with pLDDT (75.25) and PAE (10.96). These results suggest that the sequences produced by both models are structurally stable and more likely to fold into valid conformations, highlighting the effectiveness of their design modules in capturing the structural constraints of proteins.
</li>
<li>
<strong>
ProDVa and Pinal exhibit comparable performance in Language Alignment, whereas the remaining baselines demonstrate substantially inferior results.
</strong>
For language alignment, both models outperform all baselines by large margins. ProDVa achieves the best retrieval accuracy (66.83), while Pinal attains a comparable score (63.43). They also perform significantly better in ProTrek Score and EvoLlama Score compared to ProteinDT, Chroma, and PAAG. This indicates that the semantic and evolutionary information embedded in the descriptions are effectively translated into protein sequences by ProDVa and Pinal, whereas the baseline methods fail to capture such alignment.
</li>
<li>
<strong>
ProDVa and Pinal perform poorly with respect to novelty and diversity.
</strong>
While excelling in plausibility, foldability, and alignment, both models show relatively low novelty and diversity compared with baselines. The low novelty scores (ProDVa Seq/Struct: 14.64/36.31, Pinal Seq/Struct: 43.82/17.23) suggest that the designed sequences tend to remain close to the natural protein landscape. At the same time, their low diversity scores (ProDVa Seq/Struct: 83.29/36.92; Pinal Seq/Struct: 82.96/72.73) indicate that the models may confine functional design to narrow clusters in sequence/structure space. This reflects a trade-off: in order to achieve better functional alignment, ProDVa and Pinal may sacrifice exploration of diverse solutions, thereby limiting their coverage of the broader protein landscape.
</li>
</ul>
</div>
<div class="table-wrapper">
<table class="js-sort-table" id="tab-main-res-desc">
<caption>Table 5: Sequence-based results on the description-guided task</caption>
<thead>
<tr>
<th rowspan="2" class="js-sort-number">Models</th>
<th colspan="3" class="js-sort-number">Perplexity</th>
<th colspan="3" class="js-sort-number">Repetitveness</th>
<th colspan="4" class="js-sort-number">Foldability</th>
</tr>
<tr>
<th class="js-sort-number">PPL-ProtGPT2 ↓</th>
<th class="js-sort-number">PPL-ProGen ↓</th>
<th class="js-sort-number">PPL-RITA ↓</th>
<th class="js-sort-number">Repeat ↓</th>
<th class="js-sort-number">Rep-2 ↓</th>
<th class="js-sort-number">Rep-5 ↓</th>
<th class="js-sort-number">pLDDT ↑</th>
<th class="js-sort-number">% > 70 ↑</th>
<th class="js-sort-number">PAE ↓</th>
<th class="js-sort-number">% < 10 ↑</th>
</tr>
</thead>
<tbody>
<tr style="background-color: #f7f7f7">
<td>Natural</td>
<td>318.15</td>
<td>5.99</td>
<td>5.52</td>
<td>1.99</td>
<td>44.49</td>
<td>0.25</td>
<td>80.64</td>
<td>81.16</td>
<td>9.20</td>
<td>65.64</td>
</tr>
<tr>
<td>Random (U)</td>
<td>2484.04±4.53</td>
<td>21.71±0.00</td>
<td>22.14±0.01</td>
<td>0.72±0.01</td>
<td>34.59±0.03</td>
<td>0.01±0.00</td>
<td>22.96±0.04</td>
<td>0.16±0.04</td>
<td>24.85±0.01</td>
<td>0.56±0.03</td>
</tr>
<tr>
<td>Random (E)</td>
<td>3136.88±4.17</td>
<td>18.68±0.00</td>
<td>19.04±0.00</td>
<td>1.15±0.01</td>
<td>40.99±0.01</td>
<td>0.01±0.00</td>
<td>25.77±0.03</td>
<td>0.19±0.06</td>
<td>24.71±0.01</td>
<td>0.60±0.03</td>
</tr>
<tr>
<td>ProteinDT</td>
<td>1576.23±4.32</td>
<td>12.41±0.01</td>
<td>12.44±0.01</td>
<td>6.83±0.10</td>
<td>62.47±0.14</td>
<td>2.82±0.05</td>
<td>38.29±0.04</td>
<td>0.98±0.17</td>
<td>25.13±0.02</td>
<td>0.40±0.09</td>
</tr>
<tr>
<td>Chroma</td>
<td id="third">1370.21±1.48</td>
<td id="third">12.19±0.00</td>
<td id="third">12.42±0.01</td>
<td id="third">2.59±0.02</td>
<td id="third">55.41±0.03</td>
<td id="second">0.60±0.01</td>
<td id="third">59.18±0.09</td>
<td id="third">20.17±0.23</td>
<td id="third">15.03±0.04</td>
<td id="third">28.62±0.62</td>
</tr>
<tr>
<td>PAAG</td>
<td>2782.70±9.63</td>
<td>17.84±0.01</td>
<td>18.05±0.02</td>
<td id="second">2.34±0.02</td>
<td id="second">45.83±0.03</td>
<td id="first">0.09±0.01</td>
<td>28.39±0.07</td>
<td>0.07±0.03</td>
<td>25.38±0.01</td>
<td>0.10±0.03</td>
</tr>
<tr>
<td>Pinal</td>
<td id="first">308.97±0.68</td>
<td id="first">5.81±0.02</td>
<td id="first">5.78±0.02</td>
<td>12.83±0.13</td>
<td>58.26±0.16</td>
<td>4.73±0.06</td>
<td id="second">75.25±0.19</td>
<td id="second">68.93±0.33</td>
<td id="second">10.96±0.10</td>
<td id="second">58.41±0.38</td>
</tr>
<tr>
<td>ProDVa</td>
<td id="second">415.64±7.40</td>
<td id="second">7.63±0.09</td>
<td id="second">8.83±0.17</td>
<td id="first">1.92±0.05</td>
<td id="first">35.65±0.15</td>
<td id="third">2.81±0.13</td>
<td id="first">76.84±0.17</td>
<td id="first">76.27±0.59</td>
<td id="first">8.67±0.05</td>
<td id="first">67.65±0.43</td>
</tr>
</tbody>
</table>
</div>
<div class="table-wrapper">
<table class="js-sort-table" id="results-desc-stc">
<caption>Table 6: Structure-based results on the description-guided task</caption>
<thead>
<tr>
<th rowspan="2">Models</th>
<th colspan="2">Model-based Alignment</th>
<th colspan="9">Retrieval-based Alignment</th>
</tr>
<tr>
<th>ProTrek Score ↑</th>
<th>EvoLlama Score ↑</th>
<th>Soft(4) ↑</th>
<th>Soft(10) ↑</th>
<th>Soft(20) ↑</th>
<th>Normal(4) ↑</th>
<th>Normal(10) ↑</th>
<th>Normal(20) ↑</th>
<th>Hard(4) ↑</th>
<th>Hard(10) ↑</th>
<th>Hard(20) ↑</th>
</tr>
</thead>
<tbody>
<tr style="background-color: #f7f7f7">
<td>Natural</td>
<td>27.00</td>
<td>60.33</td>
<td>98.50</td>
<td>97.17</td>
<td>96.09</td>
<td>93.72</td>
<td>89.01</td>
<td>85.11</td>
<td>43.23</td>
<td>33.20</td>
<td>29.78</td>
</tr>
<tr>
<td>Random(U)</td>
<td>1.03±0.04</td>
<td>36.22±0.07</td>
<td>28.09±0.83</td>
<td>12.62±0.29</td>
<td>6.94±0.20</td>
<td>28.97±0.38</td>
<td>12.83±0.52</td>
<td>7.16±0.29</td>
<td>25.95±0.52</td>
<td>10.57±0.40</td>
<td>5.38±0.39</td>
</tr>
<tr>
<td>Random(E)</td>
<td>1.04±0.05</td>
<td>34.11±0.10</td>
<td>28.35±0.62</td>
<td>12.83±0.51</td>
<td>6.73±0.66</td>
<td>28.97±0.62</td>
<td>12.59±0.29</td>
<td>6.84±0.39</td>
<td>25.79±0.45</td>
<td>10.39±0.67</td>
<td>5.46±0.45</td>
</tr>
<tr>
<td>ProteinDT</td>
<td>1.20±0.06</td>
<td id="third">40.57±0.05</td>
<td id="third">42.91±0.68</td>
<td id="third">24.97±1.33</td>
<td id="third">16.77±1.16</td>
<td id="third">34.58±0.99</td>
<td id="third">16.56±0.44</td>
<td id="third">9.43±0.33</td>
<td>25.09±1.17</td>
<td>10.38±0.55</td>
<td>5.01±0.43</td>
</tr>
<tr>
<td>Chroma</td>
<td id="third">2.10±0.02</td>
<td>40.10±0.23</td>
<td>29.54±0.59</td>
<td>13.43±0.18</td>
<td>7.41±0.22</td>
<td>29.63±0.58</td>
<td>13.26±0.50</td>
<td>7.44±0.21</td>
<td id="third">25.51±0.47</td>
<td id="third">10.68±0.43</td>
<td id="third">5.73±0.19</td>
</tr>
<tr>
<td>PAAG</td>
<td>1.29±0.04</td>
<td>34.39±0.18</td>
<td>33.33±0.25</td>
<td>15.24±0.33</td>
<td>8.27±0.20</td>
<td>29.63±0.70</td>
<td>12.83±0.14</td>
<td>6.87±0.17</td>
<td>25.19±0.65</td>
<td>10.13±0.34</td>
<td>4.96±0.23</td>
</tr>
<tr>
<td>Pinal</td>
<td id="first">17.50±0.09</td>
<td id="first">53.40±0.31</td>
<td id="second">82.42±0.45</td>
<td id="second">74.44±0.63</td>
<td id="second">69.99±0.67</td>
<td id="second">71.69±0.59</td>
<td id="second">63.53±0.24</td>
<td id="second">58.43±0.48</td>
<td id="first">29.51±0.31</td>
<td id="first">17.89±0.26</td>
<td id="first">14.17±0.27</td>
</tr>
<tr>
<td>ProDVa</td>
<td id="second">17.40±0.06</td>
<td id="second">51.19±0.17</td>
<td id="first">85.64±0.06</td>
<td id="first">77.37±0.44</td>
<td id="first">72.75±0.61</td>
<td id="first">77.52±0.28</td>
<td id="first">66.67±0.37</td>
<td id="first">59.03±0.44</td>
<td id="second">27.84±0.71</td>
<td id="second">15.77±0.29</td>
<td id="second">12.11±0.28</td>
</tr>
</tbody>
</table>
</div>
<div class="table-wrapper">
<table class="js-sort-table" id="res-desc-alt">
<caption>Table 7: Language Alignment on the description-guided task</caption>
<thead>
<tr>
<th rowspan="2">Models</th>
<th colspan="5">Similarity</th>
<th colspan="4">Novelty</th>
<th colspan="2">Diversity</th>
</tr>
<tr>
<th>GT-Identity ↑</th>
<th>GT-TMScore ↑</th>
<th>ESM-F1 ↑</th>
<th>ESM-Precision ↑</th>
<th>ESM-Recall ↑</th>
<th>Seq<sub>Easy</sub> ↑</th>
<th>Seq<sub>Hard</sub> ↑</th>
<th>Struct<sub>Easy</sub> ↑</th>
<th>Struct<sub>Hard</sub> ↑</th>
<th>Seq ↑</th>
<th>Struct ↑</th>
</tr>
</thead>
<tbody>
<tr style="background-color: #f7f7f7">
<td>Natural</td>
<td>100.00</td>
<td>100.00</td>
<td>100.00</td>
<td>100.00</td>
<td>100.00</td>
<td>36.11</td>
<td>4.90</td>
<td>38.51</td>
<td>13.56</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td>Random(U)</td>
<td>0.37±0.03</td>
<td>16.95±0.03</td>
<td>71.06±0.02</td>
<td>81.66±0.02</td>
<td>63.46±0.02</td>
<td>98.77±0.03</td>
<td>58.14±0.07</td>
<td>96.82±0.03</td>
<td>77.64±0.12</td>
<td>97.01</td>
<td>81.59</td>
</tr>
<tr>
<td>Random(E)</td>
<td>0.23±0.04</td>
<td>17.10±0.00</td>
<td>71.95±0.02</td>
<td>82.51±0.02</td>
<td>64.35±0.02</td>
<td>98.45±0.01</td>
<td>60.19±0.14</td>
<td>96.25±0.04</td>
<td>76.82±0.10</td>
<td>99.56</td>
<td>81.45</td>
</tr>
<tr>
<td>ProteinDT</td>
<td>0.18±0.02</td>
<td id="second">13.94±0.03</td>
<td>72.80±0.05</td>
<td id="second">81.44±0.03</td>
<td>66.38±0.05</td>
<td id="third">96.92±0.12</td>
<td id="first">70.74±0.07</td>
<td id="second">94.68±0.02</td>
<td id="second">71.16±0.08</td>
<td id="first">99.23</td>
<td id="first">83.67</td>
</tr>
<tr>
<td>Chroma</td>
<td id="third">0.22±0.04</td>
<td id="third">17.93±0.02</td>
<td>72.82±0.02</td>
<td id="third">80.22±0.03</td>
<td id="third">67.06±0.01</td>
<td id="second">97.28±0.02</td>
<td id="third">58.68±0.09</td>
<td id="third">80.99±0.04</td>
<td id="third">51.06±0.21</td>
<td id="third">96.13</td>
<td id="third">79.9</td>
</tr>
<tr>
<td>PAAG</td>
<td>0.17±0.02</td>
<td>14.63±0.03</td>
<td id="third">73.26±0.03</td>
<td id="first">83.10±0.02</td>
<td>66.04±0.03</td>
<td id="first">98.90±0.02</td>
<td id="second">63.64±0.09</td>
<td id="first">96.44±0.03</td>
<td id="first">77.34±0.12</td>
<td id="second">99.15</td>
<td id="second">82.16</td>
</tr>
<tr>
<td>Pinal</td>
<td id="second">18.65±0.15</td>
<td id="first">23.75±0.14</td>
<td id="first">76.63±0.06</td>
<td>77.74±0.08</td>
<td id="first">75.99±0.06</td>
<td>55.55±0.19</td>
<td>43.82±0.22</td>
<td>40.07±0.33</td>
<td>17.23±0.23</td>
<td>82.96</td>
<td>72.73</td>
</tr>
<tr>
<td>ProDVa</td>
<td id="first">21.48±0.15</td>
<td id="second">20.03±0.11</td>
<td id="second">75.23±0.01</td>
<td>77.01±0.05</td>
<td id="second">74.11±0.02</td>
<td>38.23±0.31</td>
<td>14.64±0.23</td>
<td>56.18±23.36</td>
<td>36.31±33.02</td>
<td>83.29</td>
<td>36.92</td>
</tr>
</tbody>
</table>
</div>
</div>
<!-- Description-guided -->
</div>
</section>
<!-- Keyword-guided -->
<section class="section" id="results-keyword">
<div class="container is-max-desktop">
<div class="is-centered has-text-centered">
<h1 class="title is-3">Keyword-guided</h1>
<div class="content has-text-justified">
<p>In <a href="#results-key-seq">Table 1</a>, <a href="#results-key-stc">Table 2</a>, <a
href="#results-key-alt">Table 3</a> and <a href="#results-key-nov-div">Table 4</a>, we
report the benchmark results for keyword-guided task. Based on these results, our key findings
are as follows:</p>
<ul>
<li>
<strong>
CFP-Gen, Pinal and ESM3 show great performance in Perplexity while the Repeat show poorly.
</strong>
These models achieve the lowest perplexity scores (CFP-Gen: 4.94–5.23; Pinal: 6.85–8.12; ESM3: 6.33), indicating that their generated sequences exhibit strong rationality under the protein language model. However, they also show much higher Repeat (ranging from 11.86 to 28.13) compared with ProDVa or Chroma, suggesting that the improved plausibility comes at the cost of local redundancy in sequence design.
</li>
<li>
<strong>
CFP-Gen, ProDVa and Pinal can design foldable proteins.
</strong>
These models consistently achieve high pLDDT and low PAE across different evaluation settings. ProDVa stands out with the best overall foldability (pLDDT: 72.80–74.73; PAE: 6.11–8.06), while CFP-Gen and Pinal also produce structures with good confidence (pLDDT around 69–76; PAE around 11–14). This indicates that their design strategies are particularly effective at generating sequences that fold into stable 3D structures.
</li>
<li>
<strong>
CFP-Gen shows great performance among the keyword-guided baselines, while weak performance among the description-guided baselines.
</strong>
Compared to other keyword-guided baselines such as ProteinDT, Chroma, or PAAG, CFP-Gen achieves significantly higher alignment with biological annotations (e.g., IPR Recovery up to 35.21 and GO Recovery up to 21.05). In contrast, its performance was still less competitive than Pinal and ProDVa.
</li>
<li>
<strong>
Baselines perform better on IPR-guided task than GO-guided task.
</strong>
In the single-keyword setting, models achieve higher recovery performance on the metric that matches the input type. Moreover, using IPR as input generally leads to stronger performance across Plausibility, Foldability, and most Language Alignment metrics, indicating that IPR annotations provide more precise constraints for protein design than GO terms.
</li>
<li>
<strong>
The IPR&GO-guided task imposes stricter constraints than the other two tasks, while less than description-guided task.
</strong>
When extending from single- to dual-keyword guidance, we observe a nuanced trade-off. IPR Recovery increases while GO Recovery decreases, suggesting that IPR contributes more strongly to functional alignment in this joint setting. Meanwhile, both ProTrek Score and Retrieval Accuracy are improved, while Plausibility and Foldability remain largely unchanged. These results imply that combining IPR and GO constraints reduces the design difficulty in terms of language alignment, enabling models to better converge on functionally consistent sequences without sacrificing structural quality. Nevertheless, the alignment scores are still higher than those obtained in the description-guided setting, showing that structured keyword guidance provides clearer signals for functional targeting, albeit at the cost of reduced novelty and diversity.
</li>
</ul>
</div>
<div class="table-wrapper">
<table class="js-sort-table" id="results-key-seq">
<caption>Table 1: Plausibility and Foldability results on the keyword-guided task</caption>
<thead>
<tr>
<th rowspan="2">Models</th>
<th colspan="3">Perplexity</th>
<th colspan="3">Repetitveness</th>
<th colspan="4">Foldability</th>
</tr>
<tr>
<th>PPL-ProtGPT2 ↓</th>
<th>PPL-ProGen ↓</th>
<th>PPL-RITA ↓</th>
<th>Repeat ↓</th>
<th>Rep-2 ↓</th>
<th>Rep-5 ↓</th>
<th>pLDDT ↑</th>
<th>% > 70 ↑</th>
<th>PAE ↓</th>
<th>% < 10 ↑</th>
</tr>
</thead>
<tbody>
<tr>
<td colspan="11" style="background-color:#cac7c7">
<strong><em>guided with GO keywords</em></strong>
</td>
</tr>
<tr style="background-color: #f7f7f7">
<td>Natural</td>
<td>554.35</td>
<td>9.17</td>
<td>8.89</td>
<td>2.17</td>
<td>44.43</td>
<td>0.43</td>
<td>76.92</td>
<td>72.44</td>
<td>10.54</td>
<td>54.69</td>
</tr>
<tr>
<td>Random(U)</td>
<td>2473.84±10.48</td>
<td>21.74±0.01</td>
<td>22.18±0.02</td>
<td>0.72±0.03</td>
<td>35.52±0.04</td>
<td>0.01±0.00</td>
<td>23.20±0.02</td>
<td>0.10±0.08</td>
<td>24.56±0.01</td>
<td>0.19±0.08</td>
</tr>
<tr>
<td>Random(E)</td>
<td>3096.13±18.42</td>
<td>18.68±0.01</td>
<td>19.05±0.01</td>
<td>1.14±0.04</td>
<td>41.45±0.07</td>
<td>0.02±0.00</td>
<td>25.99±0.12</td>
<td>0.05±0.08</td>
<td>24.47±0.03</td>
<td>0.24±0.08</td>
</tr>
<tr>
<td>ProteoGAN</td>
<td>2708.39±32.50</td>
<td>18.03±0.01</td>
<td>18.31±0.02</td>
<td id="third">2.50±0.05</td>
<td id="third">42.73±0.86</td>
<td id="first">0.03±0.00</td>
<td>28.72±0.43</td>
<td>0.06±0.10</td>
<td>24.67±0.17</td>
<td>0.12±0.20</td>
</tr>
<tr>
<td>CFP-Gen</td>
<td id="first">187.72±9.71</td>
<td id="first">5.16±0.03</td>
<td id="first">4.65±0.02</td>
<td>12.67±0.79</td>
<td>59.67±0.83</td>
<td>13.82±0.74</td>
<td id="second">73.38±0.26</td>
<td id="second">65.65±1.11</td>
<td id="third">14.61±0.27</td>
<td id="third">35.20±1.76</td>
</tr>
<tr>
<td>ProteinDT</td>
<td>1531.76±17.19</td>
<td>12.23±0.06</td>
<td id="third">12.29±0.06</td>
<td>7.98±0.51</td>
<td>64.01±0.25</td>
<td>3.32±0.38</td>
<td>40.35±0.30</td>
<td>1.15±0.00</td>
<td>25.57±0.03</td>
<td>0.00±0.00</td>
</tr>
<tr>
<td>Chroma</td>
<td>1354.61±4.81</td>
<td>12.18±0.03</td>
<td>12.40±0.03</td>
<td>2.71±0.05</td>
<td>55.09±0.12</td>
<td id="third">0.67±0.03</td>
<td>59.27±0.20</td>
<td>22.17±0.65</td>
<td>15.00±0.10</td>
<td>30.93±0.08</td>
</tr>
<tr>
<td>PAAG</td>
<td>2650.36±11.01</td>
<td>18.08±0.02</td>
<td>18.38±0.02</td>
<td id="second">2.48±0.20</td>
<td id="second">39.23±0.05</td>
<td id="second">0.05±0.01</td>
<td>31.47±0.10</td>
<td>0.34±0.36</td>
<td>23.88±0.05</td>
<td>0.24±0.08</td>
</tr>
<tr>
<td>Pinal</td>
<td id="second">414.26±77.15</td>
<td id="second">6.85±0.59</td>
<td id="second">6.89±0.64</td>
<td>14.13±2.58</td>
<td>59.84±4.37</td>
<td>4.85±1.63</td>
<td id="third">72.58±5.55</td>
<td id="third">62.10±14.24</td>
<td id="second">11.79±2.52</td>
<td id="second">52.19±12.03</td>
</tr>
<tr>
<td>ProDVa</td>
<td id="third">486.77±9.51</td>
<td id="third">11.16±0.29</td>
<td>18.71±0.77</td>
<td id="first">1.87±0.07</td>
<td id="first">22.04±0.09</td>
<td>0.88±0.05</td>
<td id="first">74.73±0.24</td>
<td id="first">68.40±0.38</td>
<td id="first">6.11±0.02</td>
<td id="first">84.90±0.46</td>
</tr>
<tr>
<td colspan="11" style="background-color:#cac7c7">
<strong><em>guided with IPR keywords</em></strong>
</td>
</tr>
<tr style="background-color: #f7f7f7">
<td>Natural</td>
<td>611.99</td>
<td>9.73</td>
<td>9.47</td>
<td>2.23</td>
<td>44.05</td>
<td>0.48</td>
<td>75.77</td>
<td>68.85</td>
<td>11.13</td>
<td>50.92</td>
</tr>
<tr>
<td>Random(U)</td>
<td>2475.07±10.83</td>
<td>21.76±0.02</td>
<td>22.21±0.02</td>
<td>0.69±0.06</td>
<td>35.18±0.07</td>
<td>0.01±0.00</td>
<td>23.40±0.07</td>
<td>0.08±0.07</td>
<td>24.42±0.02</td>
<td>0.11±0.00</td>
</tr>
<tr>
<td>Random(E)</td>
<td>3104.89±24.07</td>
<td>18.67±0.02</td>
<td>19.05±0.02</td>
<td>1.16±0.02</td>
<td>40.91±0.08</td>
<td>0.02±0.01</td>
<td>26.29±0.18</td>
<td>0.08±0.07</td>
<td>24.34±0.03</td>
<td>0.19±0.07</td>
</tr>
<tr>
<td>ESM3</td>
<td id="second">330.44±9.90</td>
<td id="second">6.33±0.07</td>
<td id="second">6.59±0.07</td>
<td>28.13±0.24</td>
<td>68.98±0.42</td>
<td>21.11±0.47</td>
<td>60.90±0.77</td>
<td>32.93±2.43</td>
<td>16.73±0.27</td>
<td>22.68±1.83</td>
</tr>
<tr>
<td>CFP-Gen</td>
<td id="first">135.57±4.51</td>
<td id="first">4.94±0.12</td>
<td id="first">5.03±0.11</td>
<td>11.86±0.29</td>
<td>59.17±0.57</td>
<td>13.57±0.88</td>
<td id="first">76.36±0.35</td>
<td id="first">72.52±1.45</td>
<td id="second">12.54±0.26</td>
<td id="second">47.23±2.51</td>
</tr>
<tr>
<td>ProteinDT</td>
<td>1506.64±5.70</td>
<td>11.87±0.02</td>
<td>11.93±0.02</td>
<td>10.02±0.38</td>
<td>65.68±0.30</td>
<td>5.83±0.39</td>
<td>37.59±0.15</td>
<td>0.04±0.07</td>
<td>26.19±0.03</td>
<td>0.00±0.00</td>
</tr>
<tr>
<td>Chroma</td>
<td>1336.19±7.55</td>
<td>12.17±0.01</td>
<td>12.39±0.02</td>
<td id="third">2.60±0.08</td>
<td id="third">54.53±0.08</td>
<td id="second">0.54±0.03</td>
<td>59.76±0.26</td>
<td>23.75±1.35</td>
<td>14.67±0.05</td>
<td>31.38±0.75</td>
</tr>
<tr>
<td>PAAG</td>
<td>2748.12±25.25</td>
<td>17.85±0.04</td>
<td>18.06±0.03</td>
<td id="second">2.32±0.11</td>
<td id="second">44.78±0.06</td>
<td id="first">0.08±0.01</td>
<td>30.89±0.03</td>
<td>0.11±0.11</td>
<td>24.98±0.02</td>
<td>0.19±0.13</td>
</tr>
<tr>
<td>Pinal</td>
<td id="third">525.38±80.49</td>
<td id="third">8.12±0.45</td>
<td id="third">8.22±0.47</td>
<td>16.73±1.96</td>
<td>59.97±3.55</td>
<td>6.32±1.28</td>
<td id="third">65.69±5.42</td>
<td id="third">44.90±12.43</td>
<td id="third">14.10±2.19</td>
<td id="third">36.13±9.57</td>
</tr>
<tr>
<td>ProDVa</td>
<td>574.60±5.52</td>
<td>12.47±0.77</td>
<td>19.07±1.06</td>
<td id="first">1.99±0.02</td>
<td id="first">21.64±0.11</td>
<td id="third">1.51±0.12</td>
<td id="second">72.80±0.48</td>
<td id="second">60.65±0.65</td>
<td id="first">6.86±0.10</td>
<td id="first">79.92±1.03</td>
</tr>
<tr>
<td colspan="11" style="background-color:#cac7c7">
<strong><em>guided with IPR&GO keywords</em></strong>
</td>
</tr>
<tr style="background-color: #f7f7f7">
<td>Natural</td>
<td>534.49</td>
<td>8.96</td>
<td>8.66</td>
<td>2.16</td>
<td>45.01</td>
<td>0.44</td>
<td>77.17</td>
<td>73.15</td>
<td>10.48</td>
<td>54.9</td>
</tr>
<tr>
<td>Random(U)</td>
<td>2482.06±21.06</td>
<td>21.72±0.01</td>
<td>22.14±0.01</td>
<td>0.73±0.05</td>
<td>36.12±0.06</td>
<td>0.01±0.00</td>
<td>22.85±0.10</td>
<td>0.00±0.00</td>
<td>24.72±0.01</td>
<td>0.00±0.00</td>
</tr>
<tr>
<td>Random(E)</td>
<td>3120.95±10.54</td>
<td>18.68±0.02</td>
<td>19.03±0.01</td>
<td>1.14±0.03</td>
<td>42.08±0.02</td>
<td>0.02±0.00</td>
<td>25.60±0.06</td>
<td>0.00±0.00</td>
<td>24.59±0.02</td>
<td>0.00±0.00</td>
</tr>
<tr>
<td>CFP-Gen</td>
<td id="first">163.51±6.62</td>
<td id="first">5.23±0.04</td>
<td id="first">5.23±0.07</td>
<td>13.14±1.18</td>
<td>59.86±0.53</td>
<td>14.17±0.32</td>
<td id="second">72.70±1.07</td>
<td id="second">60.90±1.11</td>
<td id="third">14.45±0.28</td>
<td id="second">42.69±1.76</td>
</tr>
<tr>
<td>ProteinDT</td>
<td>1697.89±9.15</td>
<td>12.81±0.05</td>
<td>12.87±0.05</td>
<td>6.81±0.17</td>
<td>63.58±0.36</td>
<td id="third">2.91±0.21</td>
<td>36.46±0.31</td>
<td>0.20±0.17</td>
<td>25.75±0.06</td>
<td>0.00±0.00</td>