-
Notifications
You must be signed in to change notification settings - Fork 0
/
figures_mlcb_run_ae_clfdnn_brca.jl
158 lines (136 loc) · 11.3 KB
/
figures_mlcb_run_ae_clfdnn_brca.jl
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
include("init.jl")
include("data_processing.jl")
include("utils.jl")
include("SurvivalDev.jl")
include("mtl_engines.jl")
include("cross_validation.jl")
device!()
#### DATA
outpath, session_id = set_dirs()
infile = "Data/GDC_processed/TCGA_BRCA_surv_cf_fpkm.h5"
brca_prediction = BRCA_data(infile, minmax_norm = true)
nsamples = size(brca_prediction.samples)[1]
#clinf = assemble_clinf(brca_prediction)
#CSV.write("Data/GDC_processed/TCGA_clinical_features_survival_pam50.csv", PAM50_data )
clinf = CSV.read("Data/GDC_processed/TCGA_clinical_features_survival_pam50.csv", DataFrame)
keep = clinf[:, "clinical_data_PAM50MRNA"] .!= "NA"
y_lbls = clinf[keep, "clinical_data_PAM50MRNA"]
y_data = label_binarizer(y_lbls)
x_data = brca_prediction.data[keep,:]
#### dump processed data
outfile = h5open("Data/GDC_processed/TCGA_BRCA_fpkm_hv_norm_PAM50.h5", "w")
outfile["data"] = x_data
outfile["genes"] = brca_prediction.genes
outfile["samples"] = brca_prediction.samples[keep]
outfile["labels"] = Array{String}(y_lbls)
close(outfile)
######### TCGA breast cancer
###### Proof of concept with Auto-Encoder classifier DNN. Provides directed dimensionality reductions
## 1 CLFDNN-AE 2D vs random x10 replicat accuracy BOXPLOT, train test samples by class bn layer 2D SCATTER.
nepochs = 50_000
nfolds, ae_nb_hls, dim_redux = 5, 1, 125
brca_aeclfdnn_params = Dict("model_title"=>"AE_AE_CLF_BRCA_2D", "modelid" => "$(bytes2hex(sha256("$(now())"))[1:Int(floor(end/3))])", "dataset" => "brca_prediction",
"model_type" => "aeaeclfdnn", "session_id" => session_id, "machine_id"=>strip(read(`hostname`, String)), "device" => "$(device())",
"nsamples_train" => length(brca_prediction.samples[keep]) - Int(round(length(brca_prediction.samples) / nfolds)), "nsamples_test" => Int(round(length(brca_prediction.samples[keep]) / nfolds)),
"nsamples" => length(brca_prediction.samples[keep]) , "insize" => length(brca_prediction.genes), "ngenes" => length(brca_prediction.genes),
"nfolds" => 5, "nepochs" => nepochs, "ae_lr" => 1e-6, "wd" => 1e-3, "dim_redux" => dim_redux,
"ae_hl_size"=>128,"enc_hl_size" => 128, "dec_nb_hl" => ae_nb_hls, "dec_hl_size" => 128, "enc_nb_hl" =>ae_nb_hls, "n.-lin" => leakyrelu,
"clfdnn_lr" => 1e-6, "clfdnn_nb_hl" => 2, "clfdnn_hl_size" => 64, "outsize" => size(y_data)[2], "model_cv_complete" => false)
brca_clf_cb = dump_aeaeclfdnn_model_cb(1000, y_lbls, export_type = "pdf")
##### 2D-AE + AE + DNN (clf)
#### split 80 /20
folds = split_train_test(x_data, y_data, brca_prediction.samples;nfolds = nfolds)
model = build(brca_aeclfdnn_params; adaptative=true)
model.ae2d
# bson("test.bson", Dict("model"=> to_cpu(model)))
# model = BSON.load("model_000001000.bson")["model"]
bmodel, bmfold, outs_test, y_test, outs_train, y_train = validate_aeaeclfdnn!(brca_aeclfdnn_params, x_data, y_data, brca_prediction.samples[keep], brca_clf_cb)
X_proj = Matrix(cpu(bmodel.ae2d.encoder(bmodel.encoder(gpu(bmfold["train_x"]')))'))
tr_labels = y_lbls[bmfold["train_ids"]]
tr_embed = DataFrame(:emb1=>X_proj[:,1], :emb2=>X_proj[:,2], :cancer_type => tr_labels)
train = AlgebraOfGraphics.data(tr_embed) * mapping(:emb1,:emb2,color = :cancer_type,marker = :cancer_type) * visual(markersize =20)
fig = draw(train, axis = (;aspect = AxisAspect(1), autolimitaspect = 1, width = 1024, height =1024,
title="$(brca_aeclfdnn_params["model_type"]) on $(brca_aeclfdnn_params["dataset"]) data\naccuracy by DNN TRAIN: $(round(brca_aeclfdnn_params["clf_tr_acc"] * 100, digits=2))% TEST: $(round(brca_aeclfdnn_params["clf_tst_acc"]*100, digits=2))%"))
fig
emb_fig_outpath = "RES/$session_id/$(brca_aeclfdnn_params["model_type"])_$(brca_aeclfdnn_params["modelid"])_AE_2D_final_model_30000.pdf"
CairoMakie.save(emb_fig_outpath, fig)
bmodel, bmfold, outs_test, y_test, outs_train, y_train = validate_aeclfdnn!(brca_aeclfdnn_params, x_data, y_data, brca_prediction.samples[keep], brca_clf_cb)
## 2 CLFDNN-AE 3D vs random x10 replicat accuracy BOXPLOT, train test samples by class bn layer 3D SCATTER.
nfolds, ae_nb_hls, dim_redux = 5, 1, 3
brca_aeclfdnn_params = Dict("model_title"=>"AE_CLF_BRCA", "modelid" => "$(bytes2hex(sha256("$(now())"))[1:Int(floor(end/3))])", "dataset" => "brca_prediction",
"model_type" => "aeclfdnn", "session_id" => session_id, "machine_id"=>strip(read(`hostname`, String)), "device" => "$(device())",
"nsamples_train" => length(brca_prediction.samples[keep]) - Int(round(length(brca_prediction.samples) / nfolds)), "nsamples_test" => Int(round(length(brca_prediction.samples[keep]) / nfolds)),
"nsamples" => length(brca_prediction.samples[keep]) , "insize" => length(brca_prediction.genes), "ngenes" => length(brca_prediction.genes),
"nfolds" => 5, "nepochs" => nepochs, "ae_lr" => 1e-4, "wd" => 1e-3, "dim_redux" => dim_redux,
"ae_hl_size"=>128,"enc_hl_size" => 128, "dec_nb_hl" => ae_nb_hls, "dec_hl_size" => 128, "enc_nb_hl" =>ae_nb_hls, "n.-lin" => leakyrelu,
"clfdnn_lr" => 1e-3, "clfdnn_nb_hl" => 2, "clfdnn_hl_size" => 64, "outsize" => size(y_data)[2], "model_cv_complete" => false)
brca_clf_cb = dump_aeclfdnn_model_cb(1000, y_lbls, export_type = "pdf")
bmodel, bmfold, outs_tst, y_test, outs_tr, y_train = validate_aeclfdnn!(brca_aeclfdnn_params, x_data, y_data, brca_prediction.samples[keep], brca_clf_cb);
## 3 CLFDNN-AE in f. of bottleneck size x10 replicate BOXPLOT, train test accuracy. Random benchmark.
nfolds, ae_nb_hls = 5, 1
dim_redux_sizes = [1, 5,10,15,20,30,50,100,200,300,500,1000,2000]
for (i,dim_redux) in enumerate(dim_redux_sizes)
brca_aeclfdnn_params = Dict("model_title"=>"AE_CLF_BRCA", "modelid" => "$(bytes2hex(sha256("$(now())"))[1:Int(floor(end/3))])", "dataset" => "brca_prediction",
"model_type" => "aeclfdnn", "session_id" => session_id, "machine_id"=>strip(read(`hostname`, String)), "device" => "$(device())",
"nsamples_train" => length(brca_prediction.samples[keep]) - Int(round(length(brca_prediction.samples) / nfolds)), "nsamples_test" => Int(round(length(brca_prediction.samples[keep]) / nfolds)),
"nsamples" => length(brca_prediction.samples[keep]) , "insize" => length(brca_prediction.genes), "ngenes" => length(brca_prediction.genes),
"nfolds" => 5, "nepochs" => nepochs, "ae_lr" => 1e-4, "wd" => 1e-3, "dim_redux" => dim_redux,
"ae_hl_size"=>128,"enc_hl_size" => 128, "dec_nb_hl" => ae_nb_hls, "dec_hl_size" => 128, "enc_nb_hl" =>ae_nb_hls, "n.-lin" => leakyrelu,
"clfdnn_lr" => 1e-3, "clfdnn_nb_hl" => 2, "clfdnn_hl_size" => 64, "outsize" => size(y_data)[2], "model_cv_complete" => false)
brca_clf_cb = dump_aeclfdnn_model_cb(1000, y_lbls, export_type = "pdf")
bmodel, bmfold, outs_test, y_test, outs_train, y_train = validate_aeclfdnn!(brca_aeclfdnn_params, x_data, y_data, brca_prediction.samples[keep], brca_clf_cb);
end
###### Bottleneck dimensionality in Auto-Encoders
## 4 Auto-Encoder train, test recontstruction correlation by bottleneck size. x10 replicate BOXPLOT.
nfolds, ae_nb_hls, nepochs = 5, 1, 30_000
dim_redux_sizes = [1,2,3,4,5,10,15,20,30,50,100,200,300,500,1000,2000]
for (i,dim_redux) in enumerate(dim_redux_sizes)
brca_ae_params = Dict("model_title"=>"AE_BRCA_DIM_REDUX_$dim_redux", "modelid" => "$(bytes2hex(sha256("$(now())"))[1:Int(floor(end/3))])", "dataset" => "brca_prediction",
"model_type" => "auto_encoder", "session_id" => session_id, "machine_id"=>strip(read(`hostname`, String)), "device" => "$(device())",
"nsamples_train" => length(brca_prediction.samples) - Int(round(length(brca_prediction.samples) / nfolds)), "nsamples_test" => Int(round(length(brca_prediction.samples) / nfolds)),
"nsamples" => length(brca_prediction.samples) , "insize" => length(brca_prediction.genes), "ngenes" => length(brca_prediction.genes),
"nfolds" => 5, "nepochs" => nepochs, "mb_size" => 200, "ae_lr" => 1e-3, "wd" => 1e-4, "dim_redux" => dim_redux,
"ae_hl_size"=> 128, "enc_hl_size" => 128, "dec_nb_hl" => ae_nb_hls, "dec_hl_size" => 128, "enc_nb_hl" =>ae_nb_hls,
"nb_clinf"=>5, "model_cv_complete" => false)
brca_ae_cb = dump_ae_model_cb(1000, export_type = "pdf") # dummy
bmodel, bmfold, outs_test, x_test, outs_train, x_train = validate_auto_encoder!(brca_ae_params, brca_prediction, brca_ae_cb, clinf;build_adaptative=false);
end
fig = Figure(resolution = (1024,1024));
fig[1,1] = Axis(fig, xlabel = "steps", ylabel = "Auto-Encoder MSE loss")
steps = collect(1:100)
loss_tr = sin.(steps)
loss_tst = sin.(steps .+1)
lines!(fig[1,1], steps, loss_tst, linewidth = 4, color = (:blue,0.75))
lines!(fig[1,1], steps, loss_tr, linewidth = 4, color = (:red,0.6))
fig
###### CPHDNN noisy input features leads to overfitting problem
## no noise
for i in 1:10
brca_cphdnnclinf_params = Dict("modelid" => "$(bytes2hex(sha256("$(now())"))[1:Int(floor(end/3))])", "dataset" => "brca_prediction",
"model_type" => "cphdnnclinf", "session_id" => session_id, "machine_id"=>strip(read(`hostname`, String)), "device" => "$(device())",
"nsamples_train" => length(brca_prediction.samples) - Int(round(length(brca_prediction.samples) / nfolds)), "nsamples_test" => Int(round(length(brca_prediction.samples) / nfolds)),
"nsamples" => length(brca_prediction.samples) , "insize" => 0, "nfolds" => 5, "nepochs" =>nepochs, "wd" => 1e-3,
"nb_clinf" => 5,"cph_lr" => 1e-4, "cph_nb_hl" => 2, "cph_hl_size" => 32, "model_cv_complete" => false)
brca_cphdnnclinf_cb = dump_model_cb(1000, labs_appdf(brca_prediction.stage), export_type = "pdf")
best_model, best_model_fold, outs_test, yt_test, ye_test, outs_train, yt_train, ye_train = validate_cphdnn_clinf!(brca_cphdnnclinf_params, brca_prediction, brca_cphdnnclinf_cb, clinf)
end
## 5 CPHDNN c-index
noise_sizes = [1,2,3,4,5,10,15,20,30,50,100,200,300,500,1000,2000]
for (i, noise_size) in enumerate(noise_sizes)
brca_prediction_NOISE = BRCA_data(reshape(rand(nsamples *noise_size), (nsamples,noise_size)),brca_prediction.samples, brca_prediction.genes, brca_prediction.survt,brca_prediction.surve,brca_prediction.age, brca_prediction.stage, brca_prediction.ethnicity)
brca_cphdnnclinf_params_noise = Dict("modelid" => "$(bytes2hex(sha256("$(now())"))[1:Int(floor(end/3))])", "dataset" => "brca_prediction_noise",
"model_type" => "cphdnnclinf", "session_id" => session_id, "machine_id"=>strip(read(`hostname`, String)), "device" => "$(device())",
"nsamples_train" => length(brca_prediction.samples) - Int(round(length(brca_prediction.samples) / nfolds)), "nsamples_test" => Int(round(length(brca_prediction.samples) / nfolds)),
"nsamples" => length(brca_prediction.samples) , "insize" => noise_size, "nfolds" => 5, "nepochs" =>nepochs, "wd" => 2e-2,
"nb_clinf" => 5,"cph_lr" => 1e-4, "cph_nb_hl" => 1, "cph_hl_size" => 32, "model_cv_complete" => false)
brca_cphdnnnoise_cb = dump_model_cb(1000, labs_appdf(brca_prediction.stage), export_type = "pdf")
best_model, best_model_fold, outs_test, yt_test, ye_test, outs_train, yt_train, ye_train = validate_cphdnn_clinf_noise!(brca_cphdnnclinf_params_noise, brca_prediction_NOISE, brca_cphdnnnoise_cb, clinf)
end
###### AECPHDNN performance and overfitting by bottleneck size.
## 7 AECPHDNN C-index vs random BOXPLOT by bn size, vs CF benchmark. 2D SCATTER
###### Alternate learning and learning rate tuning in AECPHDNN
## 6 AECPHDNN C-index AE-learning rate 2D SCATTER
###### AECPHDNN performs as well as regular CPHDNN and CPH on CF
## 8 TCGA-BRCA benchmark BOXPLOT. CPHCF, CPHDNNCF, AECPHDNNCF, random.
## 9 LGN benchmark
## 10 OV benchmark