Predicting drug-target interactions and uses for drug repositioning and repurposing
Described herein are methods of predicting drug-target interactions and methods of using the information for drug repurposing. The methods described herein combine different descriptors, including, for example, atom pair similarity, shape, topology and chemical signatures, physico-chemical functional descriptors, contact points of the ligand and the target protein, chemical similarity, and docking score.
1. A method for identifying protein-drug interactions, the method comprising:
receiving test ligand molecular data corresponding to a test ligand that is a candidate drug;
receiving protein molecular data corresponding to a protein;
receiving reference ligand data corresponding to a reference ligand that binds to the protein;
generating, by a computer system, an atom pair score from an atom descriptor of the test ligand and a corresponding atom descriptor of the reference ligand;
calculating an interaction score, the interaction score including a sum comprising the atom pair score;
normalizing the interaction score using a normalization weight calculated using the formula N ∝ =1−|1−S ∝ (x)|, wherein x is a raw parameter, S(x) is a sigmoid function, and a is a tunable scalar coefficient chosen to maximize an information-preserving variance in an image of N(x);
validating, based on the interaction score exceeding a threshold value, a first interaction between the test ligand and the protein;
in response to validating the first interaction, performing an assay using the test ligand and the protein to obtain an assay result experimentally;
after validating the first interaction, receiving the assay result; and
confirming the first interaction based at least on the received assay result.
2. The method of claim 1 , further comprising:
calculating, by the computer system, a shape score including one or more shape contributions, each shape contribution corresponding to a respective shape descriptor, wherein a respective contribution includes a first part and a second part, the first part providing a first shape score from a first respective shape function of the protein and the test ligand corresponding to the respective shape descriptor, and the second part providing a second shape score from a second respective shape function of the reference ligand and the test ligand corresponding to the respective shape descriptor;
calculating, by the computer system, a total similarity score including one or more similarity contributions, each similarity contribution corresponding to a respective similarity descriptor, wherein a respective similarity contribution provides a respective similarity score between a respective similarity function of the test ligand and the respective similarity function of the reference ligand;
calculating, by the computer system, a correction score, the correction score being a difference between a first sum and a second sum, the first sum being of energies of contact points between the reference ligand and the protein; and adding the shape score, the similarity score, and the correction score to the interaction score.
3. The method of claim 2 , further comprising:
calculating a docking score between the test ligand and the protein; and adding the docking score to the interaction score.
4. The method of claim 3 , further comprising:
normalizing the interaction score using weights for the atom pair score, the first respective shape functions, the second respective shape functions, and the docking score.
5. The method of claim 3 , wherein the interaction score Z for a qth test ligand, a pth protein, and a rth reference ligand is calculated using the formula:
Z
(
q
,
p
,
r
)
=
ω
j
Y
(
p
,
q
)
+
ω
k
P
(
r
,
q
)
+
∑
m
=
1
M
[
ω
m
f
m
(
p
,
q
)
+
ω
m
′
f
m
′
(
r
,
qp
,
q
)
]
+
∑
n
=
1
N
X
n
(
r
,
q
)
+
CS
(
OLIC
)
.
6. The method of claim 2 , wherein the one or more shape contributions include a Euclidean distance metric.
7. The method of claim 2 , wherein the one or more similarity contributions include at least one of: a number of H-bond acceptors, number of H-bond donors, dipole, electron affinity, globularity, molecular weight, ClogP, number of rotatable bonds, solvent-accessible surface area, and volume.
8. The method of claim 2 , wherein:
the first sum of the correction score for a pth protein is computed as: S(OLIC−r) p =Σ n=1 NR ω n E n,p ,
where NR is a number of contact points between the reference ligand and the protein, ω n is a weighting factor for the nth contact point, and E n,p is an energy associated with the nth contact point.
9. The method of claim 2 , wherein:
the second sum of the correction score for a qth test ligand and a pth protein is computed as: S(OLIC−q) p =Σ n=1 NQ ω n E n,q,p , where NQ is a number of contact points between the test ligand and the protein, ω n is a weighting factor for the nth contact point, and E n,p,q is an energy associated with the nth contact point.
10. The method of claim 1 , wherein a portion of the reference ligand data is extracted from a known structure of a complex of the protein bound to the reference ligand.
11. The method of claim 1 , wherein calculating the atom pair score includes a similarity Tanimoto coefficient.
12. A computer product comprising a non-transitory computer readable medium storing a plurality of instructions for controlling a computer system to perform the method of claim 1 .
13. A system comprising:
the computer product of claim 12 ; and
one or more processors for executing instructions stored on the computer readable medium.
14. The method of claim 1 , wherein the assay comprises a kinase assay.
15. The method of claim 1 , wherein:
the test ligand is a first test ligand, and
the method further comprises:
calculating interaction scores for 39 additional test ligands in the same manner as for the first test ligand.
16. The method of claim 1 , wherein the threshold value is a score value.
17. The method of claim 1 , further comprising:
determining a plurality of interaction scores for a plurality of test ligands and protein combinations;
ranking the plurality of interaction scores; and
using a ranking as the threshold value.