Immerhin habe ich nicht umsonst gelernt! Wie coole Algorithmen und Schulformeln helfen, innovative Medikamente zu entwickeln

Einführung

Mein Name ist Alexander Sadovnikov . Ich bin Absolvent der ITMO- und JetBrains-Masterstudiengänge in Softwareentwicklung und leitender Entwickler von Bioinformatik-Software in der Abteilung Computational Biology von BIOCAD .

— . : .

, , , , . - . , , . , . , , . , , , , .

, , , , .

, . , , . , , . , , . , !

. ( L1, L2, L3 H1, H2, H3). : — -! . .

. , — . .

, . , , . , , , — , . , . , .

, . , , — , . , , . . , , . , .

, , . , , . , , , , Piper, .

. , . , . , , .

, , : , . , , . .

. , . 1992 - (, , , ).

. . , . , .

, :

complexes = []

#        
for rotation in rotations: 
    #        
    for shift in shifts: 
        #      
        #      
        moved_mol2 = rotate_and_shift(mol2, rotation, shift) 

        #    ,    
        #    
        new_complex = (mol1, moved_mol2) 

        complexes.append(new_complex)

#       
complexes.sort(lambda candidate_complex: score_complex(candidate_complex)) 

. , , , . — . $$$R \cdot S$ , R — , S — . $$$R \cdot S$ $$$10^5$. .

1992 , , . .

, , , . :

$$

$$\text{cell}_{ijk} = \begin{cases} 1, & \text{ } \\ p, & \text{ } \\ 0, & \text{ }. \end{cases}$$

1 , , . $$$p$, 0 1, , , . 0 . .

. , . , , . , 1. , , , . , , .

, . , , , , .

: $$$O(N^3)$, N — , $$$O(R \cdot S \cdot N^3)$. . , . , . . : $$$O(R \cdot N^6)$. , , N, $$$10^2$.

- . ( ), $$$O(N^3 \log N)$ . , FFT. $$$O(R \cdot N^3 \cdot \log N)$, , .

- :

complexes = []

#      
grid_for_mol1 = build_grid(mol1) 

#      
fft_mol1 = fft_3d(grid_for_mol1) 

#        
for rotation in rotations: 
    #       
    rotated_mol2 = rotate(mol2, rotation) 

    #       
    grid_for_mol2 = build_grid(rotated_mol2) 

    #      
    fft_mol2 = fft_3d(rotated_mol2) 

    #    .   
    #       
    # . ,  ,     
    #        
    correlations = ifft_3d(fft_mol1 * fft_mol2) 

    #    ,    
    #    
    rotation_complexes = map(lambda shift_: (mol1, shift(rotated_mol2, shift_))) 

    #         
    #     
    complexes += list(zip(rotation_complexes, correlations)) 

#       
complexes.sort(lambda (candidate_complex, score): score) 

, . , , , .

, , , . , , .

. , , , , .

, Piper, . — , , . :

$$

$$E_{SP} = \sum_{I \in N} \sum_{j \in M} SP_{ij},$$

$$$N$ $$$M$ — , $$$SP_{ij}$ — $$$i$ $$$j$ $$$SP$.

, . , , . ( ), , .

. , Piper : DARS (Decoys as Reference State) aADARS (asymmetric Antibody DARS). aADARS , DARS — .

, . aADARS, . , . — WAASP (Widely Applicable Antibody Statistical Potential). WAASP HEDGE, , :

HEDGE,

, , , , DARS aADARS.

, ( , ). 7 , !

, . , , , . -10000 , , . 10000 , .

. , : , , . , . , , . , .

. . . RMSD (Root-Mean-Square Deviation). RMSD .

, , , , . , : , , — . , . , . , .

250 , . — , . , 10000 250, . .

. , !

, . , , . . , , . , , , (, , !). , .

, , , , , . , , : — -, , ! , - , , .

, , : . , .

-, - , . -, , , , , .

, . . , , , .

, . , . , -, , , , .

. !

: , , , , — ! . , (, , ), : , .

. , — . , . , .

, , : . . , : , , , . , .

,

, . , . : - , ( ). , , . , .

. , , , . 30 , .

, !

, .

, . . , . : - !

, , , ( , ) . , , — . , , ( !) .

, , . , , . . . , , — .