A Biterm Topic Model for Sparse Mutation Data

. 2023 Mar 4;15(5):1601. doi: 10.3390/cancers15051601

Algorithm 1

Btm 2 K (V, K_{\min}, K_{\max})

1:
Input: $V \in R_{\geq 0}^{N \times M}, 1 \leq K_{\min} < K_{\max} \leq min {N, M}$
2:
Parameters: $T = number of runs for each K$
3:
$V_{big} = Samples with more than 5 times the average biterms in V$
4:
$V = The rest$
5:
for $t = 1, \dots, T$ do
6:
$V_{1}, V_{2} = split V randomly to two equal sized sets$
7:
for $k = K_{\min}, \dots, K_{\max}$ do
8:
$btm = BTM (k, V_{1} ⋃ V_{big})$
9:
$S [k, t] = btm.log-likelihood (V_{2})$
10:
$btm = BTM (k, V_{2} ⋃ V_{big})$
11:
$S [k, t] = S [k, t] + btm.log-likelihood (V_{1})$
12:
$\tilde{K} = \underset{k}{arg min} (median (S [k, :]))$
13:
repeat
14:
$K^{*} = \tilde{K}$
15:
$\tilde{K} = min {K < K^{*} | Wilcoxons-rank-sum (S [K, :], S [K^{*}, :]) > 0.05}$
16:
until $\tilde{K} < K^{*}$
17:
return $\tilde{K}$