Method and terminal for analyzing drug-disease relevance relation, non-transitory computer-readable storage medium
A terminal for analyzing a drug-disease relevance relation includes a transceiver and a processor. The transceiver is configured to receive drug information, disease information and drug-disease relevance relation information. The processor is configured to construct a pharmacodynamic relation model based on the drug information, the disease information and the drug-disease relevance relation information.
1. A terminal for analyzing a drug-disease relevance relation, comprising:
a transceiver, configured to receive drug information, disease information and drug-disease relevance relation information, wherein the drug-disease relevance relation information comprises at least one of drug-disease therapeutic relation information and drug-disease side-effect relation information; and
a processor, coupled with the transceiver and configured to construct a pharmacodynamic relation model based on the drug information, the disease information and the drug-disease relevance relation information, wherein
the pharmacodynamic relation model is configured to analyze a drug-disease relevance relation r, and the drug-disease relevance relation comprises at least one of a drug-disease therapeutic relation r 1 and a drug-disease side-effect relation r 0 , wherein
the transceiver and the processor are configured to be selected from a group consisting of:
the transceiver is configured to receive drug information; the processor is configured to analyze the drug information according to the pharmacodynamic relation model to obtain relevance relations between a drug in the drug information and different diseases, wherein the relevance relations between the drug in the drug information and different diseases comprise therapeutic relations between the drug in the drug information and different diseases and/or side-effect relations between the drug in the drug information and different diseases; and the transceiver is configured to output the relevance relations between the drug in the drug information and different diseases,
the transceiver is configured to receive disease information; the processor is configured to analyze the disease information according to the pharmacodynamic relation model to obtain relevance relations between different drugs and a disease in the disease information, wherein the relevance relations between different drugs and the disease in the disease information comprise therapeutic relations between different drugs and the disease in the disease information and/or side-effect relations between different drugs and the disease in the disease information; and the transceiver is configured to output the relevance relations between different drugs and the disease in the disease information, and
the transceiver is configured to receive drug information and disease information; the processor is configured to analyze the drug information and the disease information according to the pharmacodynamic relation model to obtain a relevance relation between a drug in the drug information and a disease in the disease information, wherein the relevance relation between the drug in the drug information and the disease in the disease information comprises a therapeutic relation between the drug in the drug information and the disease in the disease information and/or a side-effect relation between the drug in the drug information and the disease in the disease information; and the transceiver is configured to output the relevance relation between the drug in the drug information and the disease in the disease information,
the pharmacodynamic relation model comprises an evaluation function f(d,r,s) for the drug-disease relevance relation and vector space data in a real vector space, f(d,r,s)=∥M R ×dv−M R ×sv−vR∥ 2 ;
the vector space data comprises a k-dimensional column vector dv for a drug, a k-dimensional column vector sv for a disease, an n-dimensional column vector vR for the drug-disease relevance relation, and an n×k-dimensional mapping matrix M R for the drug-disease relevance relation, where d represents the drug, s represents the disease, ∥.∥2 represents L2 norm, and the n×k-dimensional mapping matrix M R for the drug-disease relevance relation comprises an n×k-dimensional mapping matrix Mc for the drug-disease therapeutic relation and/or an n×k-dimensional mapping matrix Ms for the drug-disease side-effect relation; the n-dimensional column vector vR for the drug-disease relevance relation comprises an n-dimensional column vector vr for the drug-disease therapeutic relation and/or an n-dimensional column vector vs for the drug-disease side-effect relation, and
wherein the pharmacodynamic relation is configured to provide recommendation with a relevance score of the drug-disease relevance relation, wherein a lower relevance score indicates higher significance of the drug-disease relevance relation.
2. The terminal for analyzing the drug-disease relevance relation according to claim 1 , wherein the processor is further configured to:
obtain a relevance score of the drug-disease relevance relation based on the pharmacodynamic relation model, to evaluate the drug-disease relevance relation.
3. The terminal for analyzing the drug-disease relevance relation according to claim 1 , wherein
an expression of the k-dimensional column vector dv for the drug, an expression of the k-dimensional column vector sv for the disease, an expression of the n-dimensional column vector vR for the drug-disease relevance relation, an expression of the n×k-dimensional mapping matrix M R for the drug-disease relevance relation, and an expression of the evaluation function f(d,r,s) for the drug-disease relevance relation are solved by optimizing a target function L;
L
=
∑
(
di
,
r
,
sj
)
∈
T
∑
(
di
′
,
r
,
sj
′
)
∈
T
′
max
(
0
,
f
(
di
,
r
,
sj
)
-
f
(
di
′
,
r
,
sj
′
)
+
γ
)
where
,
f
(
di
,
r
,
sj
)
=
M
R
×
dvi
-
M
R
×
svj
-
vR
2
f
(
di
′
,
r
,
sj
′
)
=
M
R
×
dvi
′
-
M
R
×
svj
′
-
vR
2
where, T represents a set with relevance, T′ represents a set without relevance, di represents a drug with relevance, sj represents a disease with relevance; dvi represents a k-dimensional column vector for the drug with correspondence, svj represents a k-dimensional column vector for the disease with correspondence; di′ represents a drug without relevance, sj′ represents a disease without relevance; dvi′ represents a k-dimensional column vector for the drug without correspondence, svj′ represents a k-dimensional column vector for the disease without correspondence; max(a,b) represents an operation of taking a maximum value of a and b, and y represents a pre-defined margin and γ≥f(di′,r,sj′)−f(di,r,sj).
4. The terminal for analyzing the drug-disease relevance relation according to claim 3 , wherein a method for optimizing the target function L is a stochastic gradient descent algorithm.
5. The terminal for analyzing the drug-disease relevance relation according to claim 3 , wherein k<n and both n and k are less than the number of types of drugs in the drug information and the number of types of diseases in the disease information.
6. The terminal for analyzing the drug-disease relevance relation according to claim 5 , wherein the dimension of the real vector space is 300-500.
7. A method for analyzing a drug-disease relevance relation, which comprises:
collecting drug information, disease information and drug-disease relevance relation information, wherein the drug-disease relevance relation information comprises at least one of drug-disease therapeutic relation information and drug-disease side-effect relation information; and
constructing a pharmacodynamic relation model based on the drug information, the disease information and the drug-disease relevance relation information, wherein
the pharmacodynamic relation model is configured to analyze a drug-disease relevance relation r, and the drug-disease relevance relation comprises at least one of a drug-disease therapeutic relation r 1 and a drug-disease side-effect relation r 0 , wherein after constructing the pharmacodynamic relation model, the method for analyzing the drug-disease relevance relation further comprises:
receiving drug information;
analyzing the drug information according to the pharmacodynamic relation model to obtain relevance relations between a drug in the drug information and different diseases, wherein the relevance relations between the drug in the drug information and different diseases comprise therapeutic relations between the drug in the drug information and different diseases and/or side-effect relations between the drug in the drug information and different diseases; and
outputting the relevance relations between the drug in the drug information and different diseases; or,
the method for analyzing the drug-disease relevance relation further comprises:
receiving disease information;
analyzing the disease information according to the pharmacodynamic relation model to obtain relevance relations between different drugs and a disease in the disease information, wherein the relevance relations between different drugs and the disease in the disease information comprise therapeutic relations between different drugs and the disease in the disease information and/or side-effect relations between different drugs and the disease in the disease information; and
outputting the relevance relations between different drugs and the disease in the disease information; or,
the method for analyzing the drug-disease relevance relation further comprises:
receiving drug information and disease information;
analyzing the drug information and the disease information according to the pharmacodynamic relation model to obtain a relevance relation between a drug in the drug information and a disease in the disease information, wherein the relevance relation between the drug in the drug information and the disease in the disease information comprises a therapeutic relation between the drug in the drug information and the disease in the disease information and/or a side-effect relation between the drug in the drug information and the disease in the disease information; and
outputting the relevance relation between the drug in the drug information and the disease in the disease information
constructing the pharmacodynamic relation model based on the drug information, the disease information and the drug-disease relevance relation information comprises:
mapping the drug information, the disease information and the drug-disease relevance relation information to a real vector space to obtain vector space data, wherein the vector space data comprises a k-dimensional column vector dv for a drug, a k-dimensional column vector sv for a disease, an n-dimensional column vector vR for the drug-disease relevance relation, and an n×k-dimensional mapping matrix M R for the drug-disease relevance relation, the n-dimensional column vector vR for the drug-disease relevance relation comprises an n-dimensional column vector vr for the drug-disease therapeutic relation and/or an n-dimensional column vector vs for the drug-disease side-effect relation, and the n×k-dimensional mapping matrix M R for the drug-disease relevance relation comprises an n×k-dimensional mapping matrix Mc for the drug-disease therapeutic relation and/or an n×k-dimensional mapping matrix Ms for the drug-disease side-effect relation;
constructing an evaluation function f(d,r,s) for the drug-disease relevance relation according to the vector space data, and f(d,r,s)=∥M R ×dv−M R ×sv−vR∥ 2 , where d represents the drug, s represents the disease, and ∥.∥2 represents L2 norm, and
wherein the pharmacodynamic relation is configured to provide recommendation with a relevance score of the drug-disease relevance relation, wherein a lower relevance score indicates higher significance of the drug-disease relevance relation.
8. The method for analyzing the drug-disease relevance relation according to claim 7 , wherein
setting a target function L according to the evaluation function f(d,r,s) for the drug-disease relevance relation,
L
=
∑
(
di
,
r
,
sj
)
∈
T
∑
(
di
′
,
r
,
sj
′
)
∈
T
′
max
(
0
,
f
(
di
,
r
,
sj
)
-
f
(
di
′
,
r
,
sj
′
)
+
γ
)
where, T represents a set with relevance, T′ represents a set without relevance, di represents a drug with relevance, sj represents a disease with relevance; dvi represents a k-dimensional column vector for the drug with correspondence, svj represents a k-dimensional column vector for the disease with correspondence; di′ represents a drug without relevance, sr represents a disease without relevance; dvi′ represents a k-dimensional column vector for the drug without correspondence, svj′ represents a k-dimensional column vector for the disease without correspondence; max(a,b) represents an operation of taking a maximum value of a and b, and y represents a pre-defined margin and γ≥f(di′,r,sj′)−f(di,r,sj);
optimizing the target function L to obtain presentation content of the vector space data; and
obtaining an expression of the evaluation function f(d,r,s) for the drug-disease relevance relation according to the presentation content of the vector space data, wherein the presentation content of the vector space data and the expression of the evaluation function f(d,r,s) for the drug-disease relevance relation constitute the pharmacodynamic relation model.
9. The method for analyzing a drug-disease relevance relation according to claim 8 , wherein a method for optimizing the target function L is a stochastic gradient descent algorithm.
10. The method for analyzing the drug-disease relevance relation according to claim 8 , wherein k<n and both n and k are less than the number of types of drugs in the drug information and the number of types of diseases in the disease information.
11. The method for analyzing the drug-disease relevance relation according to claim 10 , wherein the dimension of the real vector space is 300-500.
12. The method for analyzing the drug-disease relevance relation according to claim 8 , wherein the evaluation function f(d,r,s) for the drug-disease relevance relation comprises an evaluation function f (d,r1,s) for the drug-disease therapeutic relation and/or an evaluation function f (d,r0,s) for the drug-disease side-effect relation, wherein
f ( d,r 1, s )=∥ Mc×dv−Mc×sv−vr∥ 2
f ( d,r 0, s )=∥ Ms×dv−Ms×sv−vs∥ 2 .
13. A non-transitory computer-readable storage medium storing executable instructions that, when executed by a terminal device, cause the terminal device to:
collect drug information, disease information and drug-disease relevance relation information, wherein the drug-disease relevance relation information comprises at least one of drug-disease therapeutic relation information and drug-disease side-effect relation information; and
construct a pharmacodynamic relation model based on the drug information, the disease information and the drug-disease relevance relation information, wherein the pharmacodynamic relation model is configured to analyze a drug-disease relevance relation r, and the drug-disease relevance relation comprises at least one of a drug-disease therapeutic relation r1 and a drug-disease side-effect relation r0, wherein
the executable instructions further causes the terminal device to perform operations selected from a group consisting of:
receiving drug information; analyzing the drug information according to the pharmacodynamic relation model to obtain relevance relations between a drug in the drug information and different diseases, wherein the relevance relation between the drug in the drug information and different diseases comprise therapeutic relations between the drug in the drug information and different diseases and/or side-effect relations between the drug in the drug information and different diseases; and outputting the relevance relations between the drug in the drug information and different diseases,
receiving disease information; analyzing the disease information according to the pharmacodynamic relation model to obtain relevance relations between different drugs and a disease in the disease information, wherein the relevance relation between different drugs and the disease in the disease information comprise therapeutic relations between different drugs and the disease in the disease information and/or side-effect relations between different drugs and the disease in the disease information; and outputting the relevance relations between different drugs and the disease in the disease information, and
receiving drug information and disease information; analyzing the drug information and the disease information according to the pharmacodynamic relation model to obtain a relevance relation between a drug in the drug information and a disease in the disease information, wherein the relevance relation between the drug in the drug information and the disease in the disease information comprises a therapeutic relation between the drug in the drug information and the disease in the disease information and/or a side-effect relation between the drug in the drug information and the disease in the disease information; and outputting the relevance relation between the drug in the drug information and the disease in the disease information,
the pharmacodynamic relation model comprises an evaluation function f(d,r,s) for the drug-disease relevance relation and vector space data in a real vector space, f(d,r,s)=∥M R ×dv−M R ×sv−vR∥ 2 ;
the vector space data comprises a k-dimensional column vector dv for a drug, a k-dimensional column vector sv for a disease, an n-dimensional column vector vR for the drug-disease relevance relation, and an n×k-dimensional mapping matrix M R for the drug-disease relevance relation, where d represents the drug, s represents the disease, ∥.∥2 represents L2 norm, and the n×k-dimensional mapping matrix M R for the drug-disease relevance relation comprises an n×k-dimensional mapping matrix Mc for the drug-disease therapeutic relation and/or an n×k-dimensional mapping matrix Ms for the drug-disease side-effect relation; the n-dimensional column vector vR for the drug-disease relevance relation comprises an n-dimensional column vector yr for the drug-disease therapeutic relation and/or an n-dimensional column vector vs for the drug-disease side-effect relation, and
wherein the pharmacodynamic relation is configured to provide recommendation with a relevance score of the drug-disease relevance relation, wherein a lower relevance score indicates higher significance of the drug-disease relevance relation.
14. The non-transitory computer-readable storage medium according to claim 13 , wherein the executable instructions further cause the terminal device to:
obtain a relevance score of the drug-disease relevance relation based on the pharmacodynamic relation model, to evaluate the drug-disease relevance relation.
15. The non-transitory computer-readable storage medium according to claim 13 , wherein
an expression of the k-dimensional column vector dv for the drug, an expression of the k-dimensional column vector sv for the disease, an expression of the n-dimensional column vector vR for the drug-disease relevance relation, an expression of the n×k-dimensional mapping matrix M R for the drug-disease relevance relation, and an expression of the evaluation function f(d,r,s) for the drug-disease relevance relation are solved by optimizing a target function L;
L
=
∑
(
di
,
r
,
sj
)
∈
T
∑
(
di
′
,
r
,
sj
′
)
∈
T
′
max
(
0
,
f
(
di
,
r
,
sj
)
-
f
(
di
′
,
r
,
sj
′
)
+
γ
)
where
,
f
(
di
,
r
,
sj
)
=
M
R
×
dvi
-
M
R
×
svj
-
vR
2
f
(
di
′
,
r
,
sj
′
)
=
M
R
×
dvi
′
-
M
R
×
svj
′
-
vR
2
where T represents a set with relevance, T′ represents a set without relevance, di represents a drug with relevance, sj represents a disease with relevance; dvi represents a k-dimensional column vector for the drug with correspondence, svj represents a k-dimensional column vector for the disease with correspondence; di′ represents a drug without relevance, sj′ represents a disease without relevance; dvi′ represents a k-dimensional column vector for the drug without correspondence, svj′ represents a k-dimensional column vector for the disease without correspondence; max(a,b) represents an operation of taking a maximum value of a and b, and y represents a pre-defined margin and γ≥f(di′,r,sj′)−f(di,r,sj).
16. The non-transitory computer-readable storage medium according to claim 15 , wherein a method for optimizing the target function L is a stochastic gradient descent algorithm.
17. The non-transitory computer-readable storage medium according to claim 15 , wherein k<n and both n and k are less than the number of types of drugs in the drug information and the number of types of diseases in the disease information.