TF-MAPS/fitting_loess.R at main · lehner-lab/TF-MAPS · GitHub

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

setwd("/PATH/to/base_dir")
load("combined_muts_hnf1a.RData") #muts_hn
load("combined_muts_fg.RData") #muts_fg
load("combined_muts_fp.RData") #muts_fp

######
######## fitting the loess model
# use only core and surface residues to fit the loess model
library(dplyr)
library(rlang)
library(ggplot2)

cbPalette<- c( "#E69F00" ,    "#999999"    , "#56B4E9"   ,   "#D55E00" ,  "#009E73"    ,
 "#F0E442"   )

fit_loess_with_anchor <- function(df, binding_col, abundance_col, binding_se_col, abundance_se_col,
                                  span = 0.9, plot = TRUE) {

  # Convert to symbols
  binding_col_sym <- ensym(binding_col)
  abundance_col_sym <- ensym(abundance_col)
  binding_se_col_sym <- ensym(binding_se_col)
  abundance_se_col_sym <- ensym(abundance_se_col)

  # Filter valid data
  newdat <- df %>%
    filter(
      surf_core %in% c("core", "surf"),
      !is.na(!!binding_col_sym),
      !is.na(!!abundance_col_sym),
      !is.na(!!binding_se_col_sym),
      !is.na(!!abundance_se_col_sym)
    )

  # Create anchor points (to force curve through origin-like region)
  anchor_points <- tibble(
    !!as_string(abundance_col_sym) := c(0, -1),
    !!as_string(binding_col_sym)   := c(0, -1),
    !!as_string(binding_se_col_sym) := c(1e-5, 1e-5),
    !!as_string(abundance_se_col_sym) := c(1e-5, 1e-5),
    surf_core = "anchor"
  )

  # Combine real and anchor data
  newdat <- bind_rows(newdat, anchor_points)

  # Fit LOESS
  mylo <- loess(
    formula = as.formula(paste(as_string(binding_col_sym), "~", as_string(abundance_col_sym))),
    data = newdat,
    weights = 1 / sqrt((pull(newdat, !!binding_se_col_sym))^2),
    span = span,
    family = "symmetric"
  )

  # Predict and residuals
  df <- df %>%
    mutate(
      loess_model = predict(mylo, newdata = df),
      residual_scaled_BA = !!binding_col_sym - loess_model
    )

  # Optional diagnostic plots
  if (plot) {
    # Compute dynamic axis limits
    x_min <- min(df %>% pull(!!abundance_col_sym), na.rm = TRUE) - 0.1
    x_max <- max(df %>% pull(!!abundance_col_sym), na.rm = TRUE) + 0.1
    y_min <- min(df %>% pull(!!binding_col_sym), na.rm = TRUE) - 0.1
    y_max <- max(df %>% pull(!!binding_col_sym), na.rm = TRUE) + 0.1

    p1 <- ggplot(df, aes(x = !!abundance_col_sym, y = !!binding_col_sym)) +
      geom_point(alpha = 0.4, aes(col=surf_core)) +
      geom_line(aes(y = loess_model),  linewidth = 1,  col="blue") +
      labs(x = as_string(abundance_col_sym), y = as_string(binding_col_sym),
           title = "LOESS fit with anchor points") +
      theme_minimal() +
      coord_cartesian(xlim = c(x_min, x_max), ylim = c(y_min, y_max)) +
      geom_vline(xintercept = -0.8, linewidth=1, col="red", linetype = "dashed") + # highly destabilising variants < -0.8
      scale_colour_manual(values=cbPalette)

    p2 <- ggplot(df, aes(x = !!abundance_col_sym, y = residual_scaled_BA)) +
      geom_point(alpha = 0.4, aes(col=surf_core)) +
      geom_hline(yintercept = 0, linetype = "dashed") +
      labs(x = as_string(abundance_col_sym), y = "Residuals (Binding - LOESS fit)",
           title = "Residual diagnostics") +
      theme_minimal() +
      scale_colour_manual(values=cbPalette) +
      geom_vline(xintercept = -0.8, linewidth=1, col="red", linetype = "dashed") # highly destabilising variants < -0.8

    print(p1)
    print(p2)
  }

  invisible(list(data = df, model = mylo))
  return(df)
}

#### example run

hn_result <- fit_loess_with_anchor(
  muts_hn,
  binding_col = scaled_fitness_b1h,
  abundance_col = mean_scaled_ss_fitness,
  binding_se_col = scaled_se_b1h,
  abundance_se_col = mean_scaled_ss_se,
  plot = TRUE
)


fg_result <- fit_loess_with_anchor(
  muts_fg,
  binding_col = scaled_b1h,
  abundance_col = mean_scaled_ss_fitness,
  binding_se_col = scaled_b1h_se,
  abundance_se_col = mean_scaled_ss_se,
  plot = TRUE
)

fp_result <- fit_loess_with_anchor(
  muts_fp,
  binding_col = scaled_b1h,
  abundance_col = mean_scaled_ss_fitness,
  binding_se_col = scaled_b1h_se,
  abundance_se_col = mean_scaled_ss_se,
  plot = TRUE
)


# plot the loess
# Optional diagnostic plots


}

invisible(list(data = df, model = mylo))