98 lines (98 with data), 9.2 kB

3


E\@Ń@sx
dZd
dl
Z
d
dlZd
dlZd
dljZd
dljjZd
dl	jj
Zd
dlj
Z
d
dlmZ
d
dlmZmZ
d
dlmZ
d
dlmZ
d
dlZd
dljZd
dlZd
dlZd
dlmZ
d
dlZd
dlm Z 
d
d	l!m"Z"
d
dl#Z#d
dl$Z$d
dl%Z&d
dl'Z'd
d
l(m(Z(
d
dl)Z)d
dl*Z*dZ+dZ,d
Z-d
Z.dZ/Gddädej0âZ1ddäZ2ddäZ3ddäZ4ddäZ5ddäZ6ddäZ7ddäZ8dd äZ9d!d"äZ:dS)#z
@author: Zhi Huang
ÚN)
┌
DataLoader)┌datasets┌
transforms)
┌Variable)
┌Counter)
┌RandomOverSampler)
┌logrank_test)
┌concordance_index)
┌tqdmÚ9ÚÚ
Úcs$eZ
dZçf
d
däZddäZçZS)┌SALMONc	sótt
|âjâ
d
}d}|
tkrRtjtj|
|âtjââ|_tjtj||âtjââ|_	|
t
krÄtjtj|
|âtjââ|_tjtj||âtjââ|_	|
tt
krýtjtjt|âtjââ|_tjtjt
|âtjââ|_tjtj|||âtjââ|_	|
tt
tt
kÉ
rftt
}}tjtjt|âtjââ|_tjtjt
|âtjââ|_tjtj|||||âtjââ|_	|
tt
tt
tkÉ
rţtt
t}}}tjtjt|âtjââ|_tjtjt
|âtjââ|_tjtj||||||âtjââ|_	|
tt
tkÉr2tt
t}}}tjtj||||âtjââ|_	|
tt
tkÉr×t}tjtjt|âtjââ|_tjtjt
|âtjââ|_tjtj||||âtjââ|_	dS)NÚÚ)┌superr┌__init__┌
len_of_RNAseq┌nnZ
Sequential┌LinearZSigmoid┌encoder1┌
classifier┌len_of_miRNAseq┌encoder2┌
len_of_cnv┌
len_of_tmb┌len_of_clinical)	┌self┌	input_dim┌dropout_rate┌	label_dimZhidden1Zhidden2Z
hidden_cnvZ
hidden_tmbZhidden_clinical)
┌	__class__ę˙./home/zhihuan/Documents/SALMON/model/SALMON.pyr'sB














&




"

"




zSALMON.__init__cCsä|
jd
}d}|t
kr.|j|
â
}|j|â
}|}|tkrN|j|
â
}|j|â
}|}|t
tkr┤|j|
ddůdt
ůfâ
}|j|
ddůt
důfâ
}|jtj||fd
ââ
}tj||fd
â}|t
ttt	kÉ
r>|j|
ddůdt
ůfâ
}|j|
ddůt
t
tůfâ
}|jtj|||
ddůt
tdůffd
ââ
}tj||fd
â}|t
ttt	t
kÉ
r╠|j|
ddůdt
ůfâ
}|j|
ddůt
t
tůfâ
}|jtj|||
ddůt
tdůffd
ââ
}tj||fd
â}|tt	t
kÉ
r˘|j|
â
}tjdg
â
}|t
tt
kÉrz|j|
ddůdt
ůfâ
}|j|
ddůt
t
tůfâ
}|jtj|||
ddůt
tdůffd
ââ
}tj||fd
â}|||fS)Nr
r
)┌shaperrrrr┌torch┌catrrr┌FloatTensor)r┌
xrZx_dZcode1┌lbl_pred┌codeZcode2r#r#r$┌forwardQsD




















,



,






,
zSALMON.forward)┌__name__┌
__module__┌__qualname__rr,┌
__classcell__r#r#)
r"r$r&s
*rcCs6|jd
â
d
j
|
â
}|j|
â
jâ}|jâ}|t|
â
S)Nr
)┌maxZtype_as┌eqZdouble┌sum┌len)┌output┌labelsZpreds┌correctr#r#r$┌accuracy}s



r8cCsd|jâj
âjdâ
}tj|â
}tjt|â
g
tdŹ}d
|||k<|
jjâj
â}
tj	||
kâ
}|t|
â
S)Nr
)
┌dtypeÚ    )
┌cpu┌numpy┌reshape┌np┌median┌zerosr4┌int┌datar3)┌hazardsr6┌hazardsdatar?┌hazards_dichotomizer7r#r#r$┌accuracy_coxâs






rFc
CsĄ|jâj
âjdâ
}tj|â
}tjt|â
g
tdŹ}d
|||k<|jjâj
âjdâ
}|dk}|
jjâj
â}
||}||}|
|}	|
|}
t	|||	|
dŹ}|j
}|S)Nr
)
r9r
)Zevent_observed_AZevent_observed_Br:r:)r;r<r=r>r?r@r4rArBrZp_value)
rCr6┌survtime_allrDr?rEZidxZT1ZT2ZE1ZE2Zresults┌pvalue_predr#r#r$┌cox_log_rankŹs
















rIcCs░|
jj
âjâ}
d
}d
}|
jd}tj|
tdŹ}
xxt|â
D]l}|
|dkr8xZt|â
D]N}||||krR|d}||||krł|d}qR||||krR|d}qRWq8W||S)Ngr
)
r9r
gÓ?)rBr;r<r%r>Zasarray┌bool┌range)rCr6rGZconcordZtotalZN_test┌
i┌
jr#r#r$┌CIndexŁs













rNcCs.|
jj
âjâ}
|j
âjâjdâ
}t|||
âS)Nr
r:)rBr;r<r=r	)rCr6rGr#r#r$┌CIndex_lifelineşs


rOcCs"tj
tjtj||
â
d
â
â
}|S)NÚ)r&Zsqrtr3┌abs)┌
a┌
b┌lossr#r#r$┌frobenius_norm_loss▓s

rUc)Csť|
|d
}|
|d}|
|d}tj
|â
}	tj|â
}
tj
|â
}t|	|ddddŹ}t|
|ddddŹ}
t||ddddŹ}d}d}d}d}d}|jâ
d}É
xét||
|âDÉ
]p\}}}|}t|â
}t|â
}|rń|jâ}|jâ}|jâ}||â
\}}}|dkÉ
r|}|}|}|}n8tj||gâ
}tj||gâ
}tj||gâ
}tj||gâ
}t	|â
}t
j||gtd	Ź}x>t
|â
D]2}x*t
|â
D]}||||k|||f<É
qxWÉ
qjWtj
|â
} t| â
} |É
r└| jâ} |}!|jd
â
}"tj|"â
}#tj|"tjtj|#| dd
Źâ
|!jââ
}$||$jjâ}|d7}qĘW|jjâjâ}%t|j|â}&t|j||â}'t|j||â}(|dkÉrxtdj||âdj|(|'ââ
|%||&|'|(|jjâjâjdâ
|
|fS)Nr)┌
e┌
tr
TF)┌
batch_size┌num_workers┌
pin_memory┌shuffler
)
r9)
┌dimz
[{:s}]		loss (nn):{:.4f}z c_index: {:.4f}, p-value: {:.3e}r:r:)r&r(┌
LongTensorr┌eval┌zipr┌cudar'r4r>r@rArKr=┌exp┌mean┌logr3┌floatrB┌itemr;r<rFrIrO┌print┌format))┌modelr┌whichsetrXr`┌verboser)rVrW┌
X┌OS_event┌OS┌
dataloader┌	lblloader┌OSloader┌lbl_pred_all┌lbl_allrG┌
code_final┌loss_nn_sum┌iterrB┌lbl┌survtime┌graphr5r+r*┌current_batch_lenZ
R_matrix_testrLrMZtest_RZtest_ystatus┌theta┌	exp_theta┌loss_nn┌code_final_4_original_data┌acc_testrH┌c_indexr#r#r$┌testÂst







































$









,











rÇc

Cs"t|â
t
jkr|jjjd
dâ
dS)Nr
gÓ?)┌typerrZweightrBZnormal_)
┌
mr#r#r$┌init_weights÷s

râc	>	CsD|d
d}	|d
d}
|d
d}|	jd}t
j|	â
}
t
j|
â
}t
j|â
}t|
|ddddŹ}t||ddddŹ}t||ddddŹ}dt_t
jjd	â

t
j	d	â

t
jd	â

t||dd
Ź}|r┬|jâ
t
j
j|jâ|ddŹ}i}g|d
<g|d
<g}g}g}g}d}d}Éx*tt|
â
â
DÉ]}|jâ
d}d}d}d} d}!d}"tjâ
É
x■t|||âDÉ
]ý\}#}$}%|jâ
|#}&|É
rł|jâ}|&jâ}&|$jâ}$||&â
\}'}(})|"dkÉ
r▓|)}|%}|$}|(} n8t
j||)gâ
}t
j||$gâ
}t
j||%gâ
}t
j| |(gâ
} t|%â
}*tj|*|*gtdŹ}+x>t|*â
D]2},x*t|*â
D]}-|%|-|%|,k|+|,|-f<ÉqWÉqWt
j|+â
}.|Ér\|.jâ}.|$}/|)jdâ
}0t
j|0â
}1t
j|0t
jt
j |1|.ddŹâ
|/j!ââ
}2d}3x>|jâD]2}4|3dkÉr╠t
j"|4â
j â}3n|3t
j"|4â
j â}3Éq«W|2||3}5|dkÉrt#d|2||3fâ

|!|2j$j%â}!|5j&â
|j'â
|"d7}"t
jj(â
É
qRW| j$j)âj*â}6|Ésf||
dkÉ
rt+|j$|â}7t,|j$||â}8t-|j$||â}9|d
j.|9â

|9|kÉr«|9}|6}|dkÉrđt#dj/|!â
dj/|9|8ââ
|j.|8â

|j.|9â

|j.|!â

|j.|7â

d
}:t0|||:|||â\};}!}<}8}=}}}|d
j.|=â

É
qW|||||||fS)N┌trainr)rVrWr
TF)rXrYrZr[iÜ)
r!r
)┌lrZweight_decayrÇ)
r9)
r\z
loss_nn: %.4f, L1: %.4fz
[Training]	 loss (nn):{:.4f}z c_index: {:.4f}, p-value: {:.3e}r:)1r%r&r(r]r┌cudnnZ
deterministicr`Zmanual_seed_allZmanual_seed┌randomZseedr┌optimZAdamZ
parametersr
rKrä┌gcZcollectr_Z	zero_gradr'r4r>r@rAr=rarbrcr3rdrQrfrBreZbackward┌step┌empty_cacher;r<rFrIrO┌appendrgrÇ)>r┌
num_epochsrXZ
learning_rater ┌lambda_1r`Zmeasurerjr)rVrWZnodes_inrkrlrmrnrorprhZ	optimizer┌c_index_list┌loss_nn_all┌
pvalue_all┌c_index_all┌
acc_train_allZc_index_best┌code_outputZepochrqrrrGrsrtrurBrvrwrxr5r+r*ryZR_matrix_trainrLrMZtrain_RZ
train_ystatusrzr{r|Zl1_reg┌
WrTr}┌	acc_trainrHrriZcode_validationr~┌c_index_predr#r#r$rä˙s─




















































$






,































"
rä);┌__doc__┌argparserçr&┌torch.nnrZtorch.backends.cudnnZbackendsrć┌torch.nn.functional┌
functional┌
F┌torch.optimrł┌torch.utils.datar┌torchvisionrr┌torch.autogradr┌collectionsr┌pandas┌pd┌matplotlib.pyplot┌pyplot┌plt┌math┌imblearn.over_samplingr┌lifelines.statisticsrZlifelines.utilsr	┌tables┌csvr<r>┌jsonr
rë┌copyrrrrrZModulerr8rFrIrNrOrUrÇrârär#r#r#r$┌<module>sP































W
@