Prediction of Potential Drug–Disease Associations through Deep Integration of Diversity and Projections of Various Drug Features

. 2019 Aug 22;20(17):4102. doi: 10.3390/ijms20174102

Algorithm 1 DivePred algorithm for predicting the potential drug-disease associations.

Input: A drug-disease association matrix

Y \in ℜ^{N_{r} \times N_{d}}

and the drugs character matrix

X_{1} \in ℜ^{881 \times N_{r}}, X_{2} \in ℜ^{1426 \times N_{r}}, X_{3} \in ℜ^{4447 \times N_{r}}, X_{4} \in ℜ^{N_{d} \times N_{r}}

Output: Drug-disease association score matrix

F

, where

F_{i j}

is the association score for drug

r_{i}

and disease

d_{j}

Randomly initialize the elements in $F, H^{(v)}, W^{(v)}$ ( $1 \leq v \leq 4$ ) with the values between 0 and 1.
While $L (F^{(v)}, H^{(v)}, W^{(v)})$ not converged do
Fix $W^{(v)}$ and $H^{(v)}$ , along with an update for $F$ , using the rule:
$F_{i j}^{n e w} \leftarrow F_{i j} \cdot \frac{{(2 M * Y + 2 α_{3} \sum_{v = 1}^{4} H^{{(v)}^{T}} + 2 α_{4} \sum_{v = 1}^{4} A^{(v)} F + 2 a_{4} F A_{d})}_{i j}}{{(2 M * F + 8 F + 2 α_{4} \sum_{v = 1}^{4} U^{(v)} F + 2 a_{4} F U_{d} + α_{5} B)}_{i j}}$
For $v = 1$ to 4
Fix $F$ and $W^{(v)}$ , along with an update for $H^{(v)}$ , using the rule:
${(H_{i j}^{(v)})}^{n e w} \leftarrow H_{i j}^{(v)} \frac{{(2 α_{1} W^{{(v)}^{T}} X^{(v)} + 2 α_{3} F^{T})}_{i j}}{{(2 α_{1} W^{{(v)}^{T}} W^{(v)} H^{(v)} + α_{2} \sum_{w \neq v}^{4} H^{(w)} + 2 α_{3} H^{(v)})}_{i j}}$
End for
For $v = 1$ to 4
Fix $F$ and $H^{(v)}$ , along with an update for $W^{(v)}$ , using the rule:
${(W_{i j}^{(v)})}^{n e w} \leftarrow W_{i j}^{(v)} \frac{{(X^{(v)} H^{{(v)}^{T}})}_{i j}}{{(W^{(v)} H^{(v)} H^{{(v)}^{T}})}_{i j}}$
End for
End While