Linfeng Wang: Long short-term memory-based deep learning model for the discovery of antimicrobial peptides targeting Mycobacterium tuberculosis

Despite being preventable and treatable, TB continues to affect millions worldwide, with an estimated 10.8 million in 2023 alone (Global Tuberculosis Report 2024). This persistent burden is further compounded by the growing threat of drug-resistant TB strains. The emergence of multidrug-resistant and extensively drug-resistant TB (MDR/XDR-TB) significantly undermines current treatment regimens and public health responses. Multidrug-resistant tuberculosis (MDR-TB) refers to strains of Mtb that no longer respond to the two main first-line antibiotics, isoniazid (HR-TB) and rifampicin (RR-TB). In more severe cases, extensively drug-resistant TB (XDR-TB) arises, which not only resists these first-line drugs but also evades additional treatments, including key second-line antibiotics and newer drugs used as last-resort options (Meeting Report of the WHO Expert Consultation on the Definition of Extensively Drug-Resistant Tuberculosis 2020). Globally, an estimated 410 000 people developed MDR/RR-TB in 2022, with a treatment success rate of ∼60% (Global Tuberculosis Report 2024). This underscores the critical need for novel therapeutic strategies that can circumvent existing resistance mechanisms. The situation is further complicated by the presence of mixed infections (Wang et al. 2023), where patients harbour both drug-susceptible and drug-resistant strains, potentially masking resistance and contributing to treatment failure.

Against this backdrop, antimicrobial peptides (AMPs) have emerged as a promising class of therapeutics. These small, naturally occurring molecules, typically cationic and amphipathic, are key components of the innate immune system (Oliveira et al. 2021). By interacting with and disrupting microbial membranes, AMPs induce rapid microbial death. Some AMPs also act intracellularly, interfering with processes such as DNA replication, RNA transcription, or protein synthesis. Moreover, certain peptides contribute to immunomodulation by enhancing pathogen clearance through mechanisms such as autophagy induction. Their multifaceted modes of action, coupled with a lower propensity for resistance development, render AMPs highly attractive candidates for addressing the rising tide of drug-resistant TB (Oliveira et al. 2021).

In TB, the host’s ability to mount an effective AMP response is actively suppressed by Mtb. Recent studies have shown that the bacterial enzyme alanine dehydrogenase (Rv2780) depletes intracellular L-alanine, a metabolite that relieves inhibition on the NF-κB pathway via interaction with PRSS1 (Peng et al. 2024). By disrupting NF-κB activation, Mtb prevents the induction of key AMPs, including β-defensin 4 (DEFB4) (Peng et al. 2024), which plays a crucial role in limiting mycobacterial survival in vivo. These findings not only underscore the importance of AMPs in TB immunity but also highlight the need to identify novel peptides capable of bypassing or restoring these disrupted pathways. However, the pool of experimentally validated TB-active AMPs remains limited. This data scarcity poses a challenge for models to identify new targets, but can be addressed through transfer learning, a technique that allows models trained on large, general AMP datasets to be fine-tuned on smaller, TB-specific examples. By leveraging learned patterns of antimicrobial function, transfer learning enables the development of classifiers and generators tailored to Mtb, accelerating the search for novel therapeutic peptides in a context of growing drug resistance.

Read full article here