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
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
use crate::util::oarfish_types::TranscriptInfo;
use itertools::izip;
use rayon::prelude::*;
use statrs::function::gamma::ln_gamma;
use tracing::{error, warn};
pub fn binomial_probability(
interval_count: &[f32],
interval_length: &[f32],
distinct_rate: f64,
) -> Vec<f64> {
let interval_counts = interval_count;
let interval_lengths = interval_length;
let count_sum = interval_counts.iter().sum::<f32>();
const ZERO_THRESH: f64 = 1e-20;
// @zzare: Where does the magic number 709 come from?
const MAX_SCALE_COUNT: f64 = 709_f64;
if count_sum == 0.0 {
return vec![0.0; interval_counts.len()];
}
if distinct_rate == 0.0 {
return vec![0.0; interval_counts.len()];
}
let probabilities: Vec<f64> = interval_counts
.iter()
.zip(interval_lengths.iter())
.map(|(&count, &length)| {
if count == 0.0 || length == 0.0 {
0.0
} else {
if count < 0.0 || length < 0.0 || distinct_rate < 0.0 {
warn!(
"count: {:?}, length: {:?}, rate:{:?}",
count, length, distinct_rate
);
}
(count as f64) / (length as f64 * distinct_rate)
}
})
.collect();
// compute the quantities (in the numerator and denominator) that we will
// use to compute the binomial probabilities.
//###########################################################################################
// obtain the maximum value over the bins.
// NOTE: f32::max(f32::NAN, x) = x for all x != f32::NAN.
let max_count = interval_counts.iter().cloned().fold(f32::NAN, f32::max);
assert!(
!max_count.is_nan(),
"Max bin count of NAN was encountered. Please report this issue on GitHub"
);
//let complementary_count: Vec<f32> = interval_counts.iter().map(|&count_val| count_sum - count_val).collect();
//let max_comp_count = complementary_count.iter().cloned().fold(f32::NAN, f32::max);
//let max_val = max_count.max(max_comp_count as f32);
let max_val = max_count;
let interval_count_modified: Vec<f32> = interval_counts
.iter()
.map(|&count_val| {
if count_val == max_val {
MAX_SCALE_COUNT as f32
} else {
(((count_val as f64) * MAX_SCALE_COUNT) / (max_val as f64)) as f32
}
})
.collect();
let sum_vec = interval_count_modified.iter().sum::<f32>();
//###########################################################################################
//let sum_vec = count_sum;
let log_numerator1: f64 = ln_gamma(sum_vec as f64 + 1.0);
let log_denominator: Vec<f64> = interval_count_modified
.iter()
.map(|&count| ln_gamma(count as f64 + 1.0) + ln_gamma((sum_vec - count) as f64 + 1.0))
.collect();
let (log_numerator2, log_numerator3) : (Vec<f64>, Vec<f64>) = probabilities.iter().zip(interval_count_modified.iter()).map(|(&prob, &count)| {
let num2 = if prob > ZERO_THRESH { prob.ln() * (count as f64) } else { ZERO_THRESH.ln() * (count as f64) };
if num2.is_nan() || num2.is_infinite() {
error!("num2 is: {:?}", num2);
error!("prob and sum_vec and count is: {:?}\t {:?}\t {:?}", prob, sum_vec, count);
panic!("Incorrect result. multinomial_probability function provides nan or infinite values for log_numerator3");
}
let num3 = if (1.0 - prob) > ZERO_THRESH {(1.0 - prob).ln() * (sum_vec - count) as f64} else { ZERO_THRESH.ln() * (sum_vec - count) as f64};
if num3.is_nan() || num3.is_infinite() {
error!("num3 is: {:?}", num3);
error!("prob and sum_vec and count is: {:?}\t {:?}\t {:?}", prob, sum_vec, count);
panic!("Incorrect result. multinomial_probability function provides nan or infinite values for log_numerator3");
}
(num2, num3)
}).unzip();
let result: Vec<f64> = izip!(log_denominator.clone(), log_numerator2.clone(), log_numerator3.clone()).map(
|(denom, num2, num3)| {
let res = (log_numerator1 - denom + num2 +num3).exp();
if res.is_nan() || res.is_infinite(){ // || (res == 0.0 && *count != 0.0) {
let len = probabilities.len();
error!("{log_numerator1}, {denom}, {num2}, {num3}, {res}, {sum_vec}, {len}");
error!("interval_counts = {:?}", interval_count_modified);
error!("probabilities = {:?}", probabilities);
let t: Vec<f64> = interval_counts.iter().map(|b| sum_vec as f64 - *b as f64).collect();
error!("sum_vec - count = {:?}", t);
panic!("Incorrect result. multinomial_probability function provides nan or infinite values for result");
}
res
}).collect();
//==============================================================================================
//// Normalize the probabilities by dividing each element by the sum
////let normalized_prob: Vec<f64> = result.iter().map(|&prob| prob / sum).collect();
let sum = result.iter().sum::<f64>();
let normalized_prob: Vec<f64> = result
.iter()
.map(|&prob| {
let normalized = prob / sum;
if normalized.is_nan() {
error!(
"Warning: Division resulted in NaN. prob: {}, sum: {}",
prob, sum
);
error!("interval_counts = {:?}", interval_count_modified);
error!("Warning: result: {:?}", result);
panic!("prob_function, normalized_prob is not valid!");
}
normalized
})
.collect();
//new method of log values
//let log_values: Vec<f64> = izip!(log_denominator, log_numerator2, log_numerator3)
//.map(|(denom, num2, num3)| log_numerator1 - denom + num2 + num3)
//.collect();
//
//// Find the maximum log value to use for stability
//let max_log_value = log_values
// .iter()
// .cloned()
// .fold(f64::NEG_INFINITY, f64::max);
//
//// Compute the exponential sum in a numerically stable way
//let exp_sum: f64 = log_values
// .iter()
// .map(|&log_val| (log_val - max_log_value).exp())
// .sum();
//
//// Normalize the probabilities
//let normalized_prob: Vec<f64> = log_values
// .iter()
// .map(|&log_val| {
// let normalized = (log_val - max_log_value).exp() / exp_sum;
// if normalized.is_nan() {
// eprintln!("Warning: Division resulted in NaN. log_val: {}, max_log_value: {}, exp_sum: {}", log_val, max_log_value, exp_sum);
// panic!("prob_function, normalized_prob is not valid!");
// } else if normalized.is_infinite() {
// eprintln!("Warning: Division resulted in inf. log_val: {}, max_log_value: {}, exp_sum: {}", log_val, max_log_value, exp_sum);
// panic!("prob_function, normalized_prob is not valid!");
// }
// normalized
// })
// .collect();
//
//if normalized_prob.iter().any(|&prob| prob < 0.0 || prob > 1.0) {
// eprintln!("Warning: Normalized probability out of bounds. Normalized probabilities: {:?}", normalized_prob);
// panic!("prob_function, normalized_prob out of valid range!");
//}
//
//let normalized_sum = normalized_prob.iter().sum::<f64>();
//if(count_sum != 0.0 && normalized_sum == 0.0){
// panic!("warning in binomial_probability function: Numerical Instability. the count in each bin is non-zero but the resultant probability is zero.");
//}
normalized_prob
}
pub fn binomial_continuous_prob(txps: &mut Vec<TranscriptInfo>, bins: &u32, threads: usize) {
use tracing::info;
use tracing::info_span;
let _log_span = info_span!("binomial_continuous_prob").entered();
info!("computing coverage probabilities");
rayon::ThreadPoolBuilder::new()
.num_threads(threads)
.build()
.unwrap();
txps.par_iter_mut().enumerate().for_each(|(_i, t)| {
let temp_prob: Vec<f64> = if *bins != 0 {
/*
let bin_counts: Vec<f32>;
let bin_lengths: Vec<f32>;
let _num_discarded_read_temp: usize;
let _bin_coverage: Vec<f64>;
(
bin_counts,
bin_lengths,
_num_discarded_read_temp,
_bin_coverage,
) = bin_transcript_normalize_counts(t, bins); //binning the transcript length and obtain the counts and length vectors
//==============================================================================================
*/
let min_cov = t.total_weight / 100.;
t.coverage_bins.iter_mut().for_each(|elem| *elem += min_cov);
let (bin_counts, bin_lengths) = t.get_normalized_counts_and_lengths();
let distinct_rate: f64 = bin_counts
.iter()
.zip(bin_lengths.iter())
.map(|(&count, &length)| (count as f64) / (length as f64))
.sum();
binomial_probability(&bin_counts, &bin_lengths, distinct_rate)
} else {
std::unimplemented!("coverage model with 0 bins is not currently implemented");
};
t.coverage_prob = temp_prob;
});
info!("done");
}