|
What can we do? Alphafold is obviously capable of (occasionally) performing miracles. So, can we separate the diamonds from the mud? |
Alphafold B-factor Alphafold models have a reliability index coded in the B-factors in the PDB file. One obvious thing to do is to see if this B-factor can be used as some kind of a filter.
I divided the Alphafold B-factors in 10 bins: 0.-10.0, 10.0-20.0, etc. And I truncated the LR score at 9, so each LR score > 9 is set to 9. All residues have an Alphafold score and a LR score, and with both mapped on ten bins, all residues have their own spot in a 10-by-10 matrix. This matrix can be printed as AF-LR and LR-AF. Both have been done below:
10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
L= 0 0 32 30448 7777331044773 494932 295987 243389 4032661211703
L= 1 3 22 15751 177013 165980 93160 74784 89870 183211 490427
L= 2 3 18 12701 136129 122910 69147 57492 72219 153796 421834
L= 3 2 31 14224 144846 127660 72659 62026 79085 171173 464479
L= 4 2 28 15143 145262 123334 72524 61621 79498 173730 473540
L= 5 5 62 33028 352619 330855 194127 157890 189682 3901091049754
L= 6 9 17 9483 93451 76827 46174 40665 53604 120698 340462
L= 7 4 16 7519 72212 59570 36238 31754 41782 92113 265934
L= 8 2 6 5216 51284 42882 25747 21771 29092 64094 191253
L=9+ 5 16 8180 86008 74667 45116 38969 50418 111722 348627
10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
A= 10. 0 3 3 2 2 5 9 4 2 5
A= 20. 32 22 18 31 28 62 17 16 6 16
A= 30. 30448 15751 12701 14224 15143 33028 9483 7519 5216 8180
A= 40. 777733 177013 136129 144846 145262 352619 93451 72212 51284 86008
A= 50.1044773 165980 122910 127660 123334 330855 76827 59570 42882 74667
A= 60. 494932 93160 69147 72659 72524 194127 46174 36238 25747 45116
A= 70. 295987 74784 57492 62026 61621 157890 40665 31754 21771 38969
A= 80. 243389 89870 72219 79085 79498 189682 53604 41782 29092 50418
A= 90. 403266 183211 153796 171173 173730 390109 120698 92113 64094 111722
A=100.1211703 490427 421834 464479 4735401049754 340462 265934 191253 348627
From whic I conclude that there is no sensible Alphafold B-factor cutoff to be found. So, we will have to resort to some good-old-fashioned brute-force.
I am implementing this now... Come back in a few days