app.py

import streamlit as st
from PIL import Image, ImageFilter
import numpy as np
import pandas as pd
from streamlit_cropper import st_cropper

# Mutation site headers
mutation_site_headers = [
    3244, 3297, 3350, 3399, 3455, 3509, 3562, 3614,
    3665, 3720, 3773, 3824, 3879, 3933, 3985, 4039,
    4089, 4145, 4190, 4245, 4298, 4349, 4402, 4455,
    4510, 4561, 4615, 4668, 4720, 4773, 4828, 4882
]

# Thresholds for each mutation site
thresholds = pd.Series({
    3244: 1.094293328, 3297: 0.924916122, 3350: 0.664586629, 3399: 0.91573613,
    3455: 1.300869714, 3509: 1.821975901, 3562: 1.178862418, 3614: 0.091557752,
    3665: 0.298697327, 3720: 0.58379781, 3773: 0.891088481, 3824: 1.145509641,
    3879: 0.81833191, 3933: 2.93084335, 3985: 1.593758847, 4039: 0.966055013,
    4089: 1.465671338, 4145: 0.30309335, 4190: 1.321615138, 4245: 1.709752495,
    4298: 0.868534701, 4349: 1.222907645, 4402: 0.58873557, 4455: 1.185522985,
    4510: 1.266797682, 4561: 1.109913024, 4615: 1.181106084, 4668: 1.408533949,
    4720: 0.714151142, 4773: 1.471959437, 4828: 0.95879943, 4882: 1.464503885
})

# === Utility functions ===

def string_to_binary_labels(s: str) -> list[int]:
    bits = []
    for char in s:
        ascii_code = ord(char)
        char_bits = [(ascii_code >> bit) & 1 for bit in range(7, -1, -1)]
        bits.extend(char_bits)
    return bits

def clean_image(img: Image.Image, min_size: int = 256) -> Image.Image:
    img = img.convert("RGB")
    if img.width < min_size or img.height < min_size:
        img = img.resize((min_size, min_size))
    img = img.filter(ImageFilter.GaussianBlur(radius=1))
    return img

def image_to_binary_labels_rgb(img: Image.Image, max_pixels: int = 256) -> list[int]:
    img = clean_image(img)
    img.thumbnail((int(np.sqrt(max_pixels)), int(np.sqrt(max_pixels))))
    img_array = np.array(img)
    flat_pixels = img_array.reshape(-1, 3)

    bits = []
    for pixel in flat_pixels:
        for channel in pixel:
            channel_bits = [(channel >> bit) & 1 for bit in range(7, -1, -1)]
            bits.extend(channel_bits)
    return bits

def binary_labels_to_rgb_image(binary_labels: list[int], width: int = None, height: int = None) -> Image.Image:
    total_pixels = len(binary_labels) // 24
    if width is None or height is None:
        side = int(np.ceil(np.sqrt(total_pixels)))
        width = height = side

    needed_pixels = width * height
    needed_bits = needed_pixels * 24
    if len(binary_labels) < needed_bits:
        binary_labels += [0] * (needed_bits - len(binary_labels))

    pixels = []
    for i in range(0, needed_bits, 24):
        r_bits = binary_labels[i:i+8]
        g_bits = binary_labels[i+8:i+16]
        b_bits = binary_labels[i+16:i+24]
        r = sum(b << (7-j) for j, b in enumerate(r_bits))
        g = sum(b << (7-j) for j, b in enumerate(g_bits))
        b = sum(b << (7-j) for j, b in enumerate(b_bits))
        pixels.append((r, g, b))

    array = np.array(pixels, dtype=np.uint8).reshape((height, width, 3))
    img = Image.fromarray(array, mode='RGB')
    return img

# === Streamlit App ===

st.title("ASCII & Binary Label Converter")
tab1, tab2, tab3 = st.tabs(["Text to Binary Labels", "Image to Binary Labels", "EF → Binary"])

# Tab 1: Text to Binary
with tab1:
    user_input = st.text_input("Enter text", value="DNA")
    if user_input:
        ascii_codes = [ord(c) for c in user_input]
        binary_labels = string_to_binary_labels(user_input)

        st.subheader("ASCII Codes")
        st.write(ascii_codes)

        st.subheader("Binary Labels per Character")
        grouped = [binary_labels[i:i+8] for i in range(0, len(binary_labels), 8)]
        for i, bits in enumerate(grouped):
            st.write(f"'{user_input[i]}' → {bits}")

        st.subheader("Binary Labels (32-bit groups)")
        groups = []
        for i in range(0, len(binary_labels), 32):
            group = binary_labels[i:i+32]
            group += [0] * (32 - len(group))
            groups.append(group + [sum(group)])

        df = pd.DataFrame(groups, columns=[str(h) for h in mutation_site_headers] + ["Edited Sites"])
        st.dataframe(df)
        st.download_button("Download as CSV", df.to_csv(index=False), "text_binary_labels.csv")

# Tab 2: Image to Binary
with tab2:
    uploaded = st.file_uploader("Upload an image (jpg/png)", type=["jpg", "jpeg", "png"])
    if uploaded:
        img = Image.open(uploaded)
        st.image(img, caption="Original", use_column_width=True)
        cropped = st_cropper(img, realtime_update=True, box_color="blue", aspect_ratio=None)
        st.image(cropped, caption="Cropped", use_column_width=True)

        max_pixels = st.slider("Max pixels to encode", 32, 1024, 256, 32)
        binary_labels = image_to_binary_labels_rgb(cropped, max_pixels=max_pixels)

        st.subheader("Binary Labels from Image")
        groups = []
        for i in range(0, len(binary_labels), 32):
            group = binary_labels[i:i+32]
            group += [0] * (32 - len(group))
            groups.append(group + [sum(group)])
        df = pd.DataFrame(groups, columns=[str(h) for h in mutation_site_headers] + ["Edited Sites"])
        st.dataframe(df)

        st.subheader("Reconstructed Image")
        recon = binary_labels_to_rgb_image(binary_labels)
        st.image(recon, caption="Reconstructed", use_column_width=True)
        st.download_button("Download CSV", df.to_csv(index=False), "image_binary_labels.csv")

# Tab 3: EF → Binary
with tab3:
    st.write("Upload an Editing Frequency CSV or enter manually:")
    st.write("**Note:** Please upload CSV files **without column headers**. Just the 32 editing frequencies per row.")
    ef_file = st.file_uploader("Upload EF CSV", type=["csv"], key="ef")
    
    if ef_file:
        # Read CSV without headers and assign mutation site headers
        ef_df = pd.read_csv(ef_file, header=None)
        ef_df.columns = [str(site) for site in mutation_site_headers]
    else:
        ef_df = pd.DataFrame(columns=[str(site) for site in mutation_site_headers])


    edited_df = st.data_editor(ef_df, num_rows="dynamic")

    if st.button("Convert to Binary Labels"):
        int_map = {str(k): k for k in thresholds.index}
        matching_cols = [col for col in edited_df.columns if col in int_map]

        binary_part = pd.DataFrame()
        for col in matching_cols:
            col_threshold = thresholds[int_map[col]]
            binary_part[col] = (edited_df[col].astype(float) >= col_threshold).astype(int)

        non_binary_part = edited_df.drop(columns=matching_cols, errors='ignore')
        binary_df = pd.concat([non_binary_part, binary_part], axis=1)

        def color_binary(val):
            if val == 1: return "background-color: lightgreen"
            if val == 0: return "background-color: lightcoral"
            return ""

        st.subheader("Binary Labels")
        styled = binary_df.style.applymap(color_binary, subset=matching_cols)
        st.dataframe(styled)
        st.download_button("Download CSV", binary_df.to_csv(index=False), "ef_binary_labels.csv")

        # Convert to bitstrings and strings
        binary_strings = []
        decoded_strings = []
        for _, row in binary_part.iterrows():
            bitlist = row.values.tolist()
            bitstring = ''.join(str(b) for b in bitlist)
            binary_strings.append(bitstring)
            decoded_strings.append(binary_labels_to_string(bitlist))

        st.subheader("Binary as Bitstrings")
        for i, b in enumerate(binary_strings):
            st.code(b)

        st.subheader("Decoded ASCII Strings")
        for i, s in enumerate(decoded_strings):
            st.write(s)